Story Studio

Can an image model draw the character a child invented, on every page of their story?

Newer version: Image judging, version 2 scores new material with four graders from four companies, four prompt roles and the evidence behind every score. This page is unchanged and still the record of the earlier runs.

Story Studio writes personalized illustrated stories for K-8 readers. Every page of every story is a generated illustration. This is the record of eight experiments that measured whether those illustrations actually do their job -- and which model and prompting recipe we should ship.

Why this existed

We were about to make decisions that assumed the pictures were fine: which model to standardize on, whether to build a new prompt-assembly stage, what to budget per story. Nobody had measured any of it. Four things have to be true for an illustrated story to work, and none of them had been tested.

1

The right character

A student builds a character, then it has to be recognisably the same character on every page of the story. Not a similar girl -- that girl, with her glasses and her jacket.

2

The same place

Consecutive pages set in one room have to look like one room, page after page, even as the camera moves and the lighting changes.

3

The right scene

The picture has to show what the paragraph actually says: the right characters, doing the right thing, with the right objects, in the right arrangement.

4

The chosen style

The student picks an art style -- graphic novel, 3D animation -- and the illustration has to actually look like that style, not an approximation of it.

How it was tested

Every experiment is a controlled comparison. Same character, same scene, same model -- change one thing and see what moves. One model grades every image, so the numbers can be compared to each other, and it returns a written reason for each score rather than a bare number. Findings that mattered were repeated against a second set of reference images before being believed, which is how one of the early conclusions got overturned.

~1,000images generated and scoredevery one judged by the same model
8experimentseach a controlled comparison
4models comparedagainst today's incumbent
$0.040per image, all-inabout what we pay now
<$30total program costrenders plus judge tokens

What we concluded

The recommendation

Ship gpt-image-2, with prompts tuned per style and a quality check on each character reference at the moment it is created. It was the strongest arm on every style tested -- 4.58 out of 5 on character identity against nano-banana's 4.00 -- and the only model to score a clean 5/5/5 on style, identity, and page-to-page consistency. At $0.040 an image with a reference attached it costs about what we already pay, so the migration is close to cost-neutral rather than a spend increase.

nano-banana is the fallback at effectively the same price and three to four times faster, blocked on a written child-content policy answer from the vendor -- a legal timeline, not an engineering one.

Stable Image Core stays, in a smaller role. It renders our styles well and cannot assemble a scene: 32% of scripted elements placed, 45% of those attached to the wrong character, and none of the four spatial relationships right. It keeps creating character references and stops rendering pages.

What it was worth

Three of the most useful outcomes are things that did not get built:

  • An engineering workstream cancelled before it was scoped. A measured 20-point accuracy gain looked like it justified building a prompt-assembly stage. Taking that prompt apart showed the whole gain came from one component production already has. The correction cost about $2.50 in renders.
  • A wrong conclusion caught before it shipped. The first style experiment concluded that heavy style language fights the reference image. Repeating it against cleaner references showed that was an artifact of the first reference set. Without the repeat, it would have become a rule in our prompt design.
  • Two live production defects found and patched -- including a missing null check that puts a null-year-old human into the character description on every page of an affected story. Neither had been reported.
  • A one-sentence fix for a defect affecting most images. Our prompt never tells the model not to draw text. Without that sentence gpt-image-2 invented lettering -- fake book titles, classroom posters, speech bubbles with dialogue nobody wrote -- in 26 of 32 images. With it: zero of 32.

How to read this site

Each tab is one experiment

They run in order, and later ones exist because earlier ones raised a question. The tab bar is directly above this page.

Each card is one generated image

Under every image sits the prompt that produced it and the judge's own written reasoning. Nothing on this site is a summary you have to take on faith -- you can check any number against the images behind it.

The colored pills are scores

Green is good, red is not: continuity5style3continuity1 Hover any pill for what it measures. Hover any underlined term for a definition, and see the Terminology & methods tab for the full list.

Rows start collapsed

Click a heading to open its images. The scores stay visible while collapsed, so a closed page still shows every score. Expand all is at the top.

Where to start

if you only look at three things

Still open: written vendor confirmation on child-content policy, review of the two production patches, integration scoping for the migration, and one measurement deliberately left undone -- whether prompt structure starts earning its keep on scenes denser than a story page currently asks for.

Story Studio style adherence -- experiment 6, multi-character

Findings
  • Describing each character separately helps. The tuned prompt identifies characters in one short clause. We split that clause in two, one per character. Of the eight combinations tested, four improved, four stayed level, and none got worse. The average gain is small (+0.25 on a 5-point scale, four images behind each number), so what matters is where the gains landed: only on combinations where one character was being crowded out by the other. Mr Penguin with Brad in graphic novel, the weakest pairing we tested, gained a full point.
  • No character swapped features with the other in any of the 96 images -- no penguin coloring on the boy, nothing of that kind. Both characters showed up in 63 of 64 images with the shared sentence and 31 of 32 with the split one. Ninety-six clean images is not proof it never happens. It does rule out the obvious worry that describing them separately would make the model mix them up.
  • Pair continuity tracks the weaker member. Brad + Aisha (both anchor-rich) held 3.75-4.25, on par with solo experiments; Mr Penguin + Brad fell to 3.00-3.50 on graphic-novel, driven by the penguin's known design ambiguity.
  • The shared sentence looks like the culprit. With both characters in one sentence, tuning stopped paying: results moved between -0.50 and +0.50, the only experiment where tuning failed to win. Splitting the sentence brought the gain back. So one sentence probably cannot describe two characters precisely enough. We only tested two pairs and two styles, so we cannot rule out a harder limit on multi-character pages.
  • gpt-image-2 met or beat nano-banana in every matched cell; 5.00 style on all 8 of its arms.
Recommended next steps
  • Check each character's reference image when it is first created, before letting two characters share a page. On what we tested, multi-character pages needed no different model -- only the change to how the prompt describes them.
  • Compose identity per character in production -- one labeled clause each, plus a sentence forbidding features from moving between them -- rather than concatenating the whole cast into a single description. It costs no extra prompt length and was worth up to a full point on the hardest pairing tested.
  • Mint the vibrant-fantasy and fantasy-art pair exemplars to finish the matrix.
Read the full written results for this experiment ↓method, every table, the reads, and the standing caveats

Experiment 3's protocol with TWO characters in every image.

  • What's tested: can a model reproduce two characters at once from one pair reference, without mixing them up.
  • Pair-mode judging: the judge checks each character's anchors separately; the weaker character's match sets the continuity score.
  • Swap watch: any feature migrating between the two (attribute swapping / blending) gets flagged explicitly.
  • Scope so far: two pairs (Mr Penguin + Brad; Brad + Aisha), two styles (3D Animation, Graphic Novel), lead model pair, both prompt arms.
  • More styles and pairs join as their exemplars are minted.
pairs Mr Penguin + Brad, Brad + Aishajudge claude-opus-4-7 via Bedrock, pair modearms sic-verbatim vs tunedruns 2026-07-31
3D Animation

3D Animation

Mr Penguin and Brad

Maria 3D Animation exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptA proportional penguin with wide,curious eyes and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters present. Mr Penguin retains black back, white belly, orange beak, orange feet, wide eyes -- solid match. Brad retains short dark hair, thick brows, brown eyes, tan jacket with dark hood/undershirt, black trousers -- identity holds. Minor drift: Brad looks slightly younger/rounder in the new image and the jacket detailing is a bit simpler, but no anchor is lost and no attribute swapping between the two.

promptA proportional penguin with wide,curious eyes and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters present. Brad: short dark wavy hair, tan jacket with dark hood/sweater underneath, black trousers -- all anchors held. Mr Penguin: black back, white belly, orange beak, wide curious eyes all preserved; orange webbed feet are barely visible/partially obscured but appear consistent. Minor drift: penguin's proportions look slightly smaller/rounder and eye styling a touch different, but identity is unmistakable. No attribute swapping between characters.

promptA proportional penguin with wide,curious eyes and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters are present. Mr Penguin retains black back, white belly, orange beak, orange webbed feet, and wide curious eyes -- all anchors intact. Brad has the same short dark hair, thick brows, tan jacket, and black trousers. Minor drift: Brad's face looks slightly younger/rounder and his hair is a bit wavier/messier than the reference's neater style, and the jacket detailing is simplified. No attribute swapping between the two.

promptA proportional penguin with wide,curious eyes and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters present. Brad retains short dark hair, thick eyebrows, tan jacket over dark hoodie, black trousers -- strong match. Mr Penguin retains black back, white belly, orange beak, wide curious eyes; orange webbed feet are less visible/mostly hidden but the visible foot appears orange-toned. Minor drift: penguin's proportions look slightly more stylized but identity holds. No attribute swapping between characters.

promptA proportional penguin with wide,curious eyes and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters present. Mr Penguin retains black back, white belly, orange beak, orange webbed feet, and wide eyes -- fully consistent. Brad retains short dark wavy hair, tan jacket with hood, and similar face structure with brown eyes. Minor drift: Brad's expression is softer/warmer and his trousers are barely visible but appear consistent. No attribute swapping between the two characters.

promptCharacter: the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters present. Mr Penguin retains black back, white belly, orange beak, orange feet, wide curious eyes -- identity intact. Brad has the same short dark hair, thick eyebrows, tan hooded jacket, and black trousers; facial structure matches closely. Minor drift: Brad's hair looks slightly less neatly styled and the jacket detailing is a touch simplified, but no attribute swapping between characters.

promptCharacter: the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters are present. Mr Penguin retains black back, white belly, orange beak, orange webbed feet, and wide eyes -- fully consistent. Brad retains short dark hair, tan jacket, and black trousers, with similar face shape and brown eyes. Minor drift: Brad's eyebrows appear less thick/prominent in the new image and his expression is more exaggerated, but identity holds. No attribute swapping between characters.

promptCharacter: the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters present. Brad retains short dark wavy hair, tan jacket, dark trousers, thick eyebrows, and similar facial structure -- strong match. Mr Penguin retains black back, white belly, orange beak, wide curious eyes; orange webbed feet are partially visible on one foot in the new image. Minor drift: penguin's proportions look slightly slimmer/taller and beak shape is a bit more pointed, but identity clearly holds. No attribute swapping between the two.

promptCharacter: the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.

gpt-image-2 -- tuned-split

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters present. Mr Penguin retains black back, white belly, orange beak, orange webbed feet, and wide curious eyes -- solid match. Brad retains short dark wavy hair, thick eyebrows, brown eyes, tan jacket with dark inner hoodie/teal shirt, and black trousers -- face shape and features match well. No attribute swapping between characters. Minor drift: Brad's hair looks slightly wavier/more voluminous than the reference's flatter style, but identity is clearly preserved.

promptCharacter 1: Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. Character 2: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters are present. Mr Penguin retains black back, white belly, orange beak, orange feet, wide curious eyes -- all anchors intact. Brad has the same short dark wavy hair, tan jacket, and black trousers, matching facial features (thick brows, brown eyes, rosy cheeks). Minor drift: Brad's inner layer/shirt is less visible and the jacket looks slightly simpler, but identity is unmistakable. No attribute swapping between characters.

promptCharacter 1: Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. Character 2: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters are present. Mr Penguin retains black back, white belly, orange beak, orange webbed feet, and wide eyes -- all anchors intact. Brad retains short dark wavy hair, tan jacket, and black trousers, with the same face shape and thick eyebrows. Minor drift: Brad's under-layer looks a bit different and his hair is slightly less textured, but no attribute swapping between characters.

promptCharacter 1: Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. Character 2: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters present. Brad retains short dark wavy hair, thick brows, tan jacket with dark hood/sweater, black trousers -- fully consistent. Mr Penguin retains black back, white belly, orange beak, orange webbed feet, and wide curious eyes. Minor drift: penguin's overall shape is a touch slimmer and the beak reads slightly darker/less vivid orange than the reference, but identity clearly holds. No attribute swapping between the two.

promptCharacter 1: Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. Character 2: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters present. Mr Penguin retains black back, white belly, orange beak, wide curious eyes -- anchors intact (feet not visible but occluded, not a break). Brad retains short dark hair, tan jacket, Latino features, and dark trousers. Minor drift: Brad's face looks slightly younger/rounder and eyes appear larger than reference, but identity is clearly preserved. No attribute swapping between the two.

promptA proportional penguin with wide,curious eyes and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Both characters present. Brad retains short dark hair, tan jacket, black trousers, backpack, and similar facial features -- strong match. Mr Penguin retains black back, white belly, wide curious eyes, but the orange beak now appears more muted/orange-tan and the orange webbed feet are barely visible/less prominent. Minor drift on penguin's beak color saturation, otherwise consistent.

promptA proportional penguin with wide,curious eyes and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Both characters are present. Mr Penguin retains black back, white belly, orange beak, orange webbed feet, and wide curious eyes -- fully intact. Brad retains short dark wavy hair, thick eyebrows, tan jacket, and dark trousers; his face shape and features read as the same boy. Minor drift: Brad now wears a knit beanie (added element) and his expression is more open-mouthed excited, but identity anchors hold. No attribute swapping between the two.

promptA proportional penguin with wide,curious eyes and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Both characters present. Mr Penguin retains black back, white belly, orange beak, orange feet, and wide curious eyes -- solid match. Brad retains short dark hair, tan jacket, dark trousers, similar face shape and expression; eyebrows slightly less bushy and hair a bit more styled, but clearly the same boy. Minor drift on Brad, no attribute swapping between characters.

promptA proportional penguin with wide,curious eyes and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity4fidelity4style5cross page consistency4

Both characters are present. Brad retains short dark hair, thick eyebrows, tan jacket layered over darker top, and black trousers with brown shoes -- identity holds well. Mr Penguin keeps black back, white belly, orange beak, and wide curious eyes; orange webbed feet are visible but partially obscured. Minor drift: Brad's hair looks slightly straighter/less wavy and the penguin's proportions are a bit taller/slimmer than the reference's chubby build, but no attribute swapping between the two.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity4style5cross page consistency4

Both characters are present. Mr Penguin retains black back, white belly, and dark beak, though the beak looks more dark/grey than clearly orange in the new image and the orange feet are less visible -- a minor drift. Brad keeps the short dark hair, tan jacket with dark collar, black trousers, and backpack -- his identity holds well. No attribute swapping between the two.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity3fidelity4style5cross page consistency4

Brad's anchors hold well: short dark hair, tan jacket, black trousers, similar face shape and expression style. Mr Penguin is present but significantly diminished -- he appears squished/blended against Brad's back like a backpack or plush, his orange beak is not clearly visible (face looks obscured or turned), and his form is distorted. Both characters are present but the penguin's identity is weak, and there's near-blending with the boy (penguin looks fused to his back). The weaker character governs, so 3.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity3fidelity5style5cross page consistency4

Both characters are present. Brad retains short dark hair, tan jacket, black trousers, and backpack -- solid match, though face shape is slightly rounder/younger. Mr Penguin keeps the black back, white belly, orange beak, and orange feet, but his proportions have shifted -- he's taller and slimmer with a more elongated body compared to the rounder, chubbier reference penguin, and his eye styling is different (more anime-like highlights). Penguin is the weaker match, dropping to a 3.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

nano-banana -- tuned-split

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned-split
continuity4fidelity4style5cross page consistency4

Both characters present. Mr Penguin retains black back, white belly, orange beak, orange webbed feet, and wide curious eyes -- solid match. Brad has the same short dark wavy hair, tan jacket, and black trousers, with a similar face shape and expression. Minor drift: Brad's jacket detailing looks slightly simplified and the teal undershirt is less visible, but identity holds for both.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. With them is Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Together they are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters present. Mr Penguin retains black back, white belly, orange beak, orange webbed feet, and wide curious eyes -- fully consistent. Brad has short dark wavy hair, tan jacket, black trousers, and backpack matching the reference; facial structure and eyes are close. Minor drift: Brad's expression/eyebrow shape is slightly more exaggerated/worried, but identity holds. No attribute swapping between the two.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. With them is Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Together they are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters are present. Mr Penguin retains black back, white belly, orange beak, orange webbed feet, and wide eyes -- fully consistent. Brad has the same short dark wavy hair, tan jacket, and black trousers, though his face reads slightly younger/more animated and his mouth is open in excitement (expected for the action). No attribute swapping between characters. Minor drift on Brad's facial proportions keeps this from a 5.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. With them is Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Together they are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned-split
continuity4fidelity4style5cross page consistency4

Both characters are present. Mr Penguin retains black back, white belly, orange beak, wide curious eyes -- webbed feet not visible in new image but that's a minor omission. Brad has the same short dark wavy hair, tan jacket, and black trousers, matching facial structure. No attribute swapping between characters. Minor drift: Brad's expression and slight facial proportions feel a touch different, but identity holds.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. With them is Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Together they are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.

Brad and Aisha

Maria 3D Animation exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters present. Aisha: long dark wavy hair, green V-neck dress, necklace, brown skin -- all preserved. Brad: short dark wavy hair, tan zip jacket, green shirt underneath, black trousers, watch on wrist -- all preserved. Minor drift: Aisha's dress lacks the visible belt from the reference, and Brad's backpack is absent (though not required in scene). No attribute swapping between them. Faces and skin tones consistent.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Both characters are present and recognizable. Brad retains short dark wavy hair, tan jacket with red patch, black trousers, and green shirt underneath. Aisha retains long dark wavy hair, green dress with brown belt, and necklace. Minor drift: Brad's hair appears slightly more voluminous/spiky, and Aisha's face is stylized with larger eyes due to the Pixar style shift, but identity anchors hold on both.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity5fidelity4style5cross page consistency4

Both characters are present and recognizably the same. Brad retains his short dark wavy hair, tan jacket, green shirt underneath, and black trousers. Aisha keeps her long dark hair, green dress with V-neck detail, belt with buckle, and necklace. No attribute swapping between them. Facial features (eye shape, eyebrows, skin tone) match closely.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters are present and clearly recognizable. Aisha retains her long dark wavy hair, green short-sleeved dress with belt, and necklace. Brad retains his short dark hair, tan jacket with red circular patch on sleeve, green shirt underneath, and black trousers. Brad's hair appears slightly more voluminous/styled than the reference's shorter cut, a minor drift. No attribute swapping between the two.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters are present and recognizable. Aisha retains her long dark wavy hair, green V-neck dress, small pendant necklace, brown eyes, and skin tone. Brad retains his short dark hair, tan jacket, green shirt underneath, black trousers, and facial features. Minor drift: Brad's hair appears slightly more voluminous/styled and Aisha's belt is not visible (obscured by book/pose), but no attribute swapping between the two.

promptCharacter: the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters are present and recognizable. Aisha retains her long black hair, green short-sleeved dress with V-neck, necklace, and similar facial features. Brad retains his short dark wavy hair, tan jacket over green shirt, and black trousers. Minor drift: Aisha's hair looks slightly shorter/less long in the new image, and Brad's face is slightly younger-looking, but no attribute swapping between characters.

promptCharacter: the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters are present. Aisha retains long dark hair, green V-neck dress with belt, earrings, and necklace -- strong match. Brad retains short dark hair, tan jacket, and dark trousers, but his green shirt under the jacket (visible in reference) is not clearly shown here, and his jacket looks slightly more shirt-like/open. No attribute swapping between the two. Minor drift on Brad's outfit details.

promptCharacter: the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters are present and clearly recognizable. Aisha retains her long dark wavy hair, green short-sleeved dress with the notched neckline, necklace, and tan skin tone. Brad retains his short dark wavy hair, tan jacket over green shirt, black trousers, and watch. No attribute swapping between the two. Minor drift: Brad's hair looks slightly longer/fluffier and Aisha's belt is not visible due to the crouching pose, but these are minor.

promptCharacter: the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.

gpt-image-2 -- tuned-split

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters are present and clearly recognizable. Aisha retains her long dark wavy hair, green dress, and necklace, though the brown belt visible in the reference is hidden by the book. Brad keeps his short dark wavy hair, tan jacket over green shirt, and dark trousers. Facial features and skin tones match well for both. No attribute swapping between the two. Minor drift: Aisha's earrings appear as small studs rather than clearly gold hoops, and the belt isn't visible.

promptCharacter 1: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Character 2: Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tuned-splitref gap
continuity4fidelity5style5cross page consistency4

Both characters are present and recognizable. Brad retains short dark wavy hair, tan jacket, and black trousers. Aisha retains long dark wavy hair, green dress with brown belt, and necklace. Minor drift: Aisha's earrings are small studs rather than clearly gold hoops, and the backpack Brad wore in the reference is absent -- but the backpack wasn't a sheet anchor. No attribute swapping between the two.

promptCharacter 1: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Character 2: Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned-splitref gap
continuity4fidelity5style5cross page consistency4

Both characters are present and clearly recognizable. Aisha retains her long black hair, green V-neck dress, and brown belt with buckle. Brad retains his short dark wavy hair, tan jacket, and black trousers. Minor drift: Aisha's gold hoop earrings are not clearly visible, and Brad's green undershirt is less apparent under the zipped jacket. No attribute swapping between characters.

promptCharacter 1: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Character 2: Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters are present and recognizable. Brad retains short dark wavy hair, tan jacket, green shirt underneath, and dark trousers. Aisha retains long dark wavy hair, green dress with brown belt, and pendant necklace. Minor drift: Aisha's earrings are less clearly gold hoops and her hair looks slightly shorter/similar length; Brad's features are close but slight facial softening. No attribute swapping between characters.

promptCharacter 1: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Character 2: Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters are present. Aisha retains dark hair, green dress, tan skin, and expressive dark eyes -- her hair appears slightly shorter/wavier than the reference's long straight hair, a minor drift. Brad retains short dark wavy hair, tan jacket with red patch, green shirt underneath, black trousers, and backpack straps -- anchors intact. No attribute swapping between the two.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Both characters present. Aisha: long dark hair, green dress, necklace, warm skin tone -- all preserved (belt less visible but dress is same). Brad: short dark wavy hair, tan jacket with red patch on sleeve, green shirt underneath, dark trousers, backpack -- all preserved. Facial features and proportions match well for both. Minor drift: Brad's hair looks slightly shorter/less voluminous, but identity is intact. No attribute swapping between the two.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters are present. Aisha retains long dark hair, green dress with belt, and similar facial features. Brad retains short dark hair, tan jacket, black trousers, and backpack. Minor drift: Aisha's dress sleeves appear shorter/different and her expression is much more animated, and Brad's jacket looks slightly more like a vest/lighter build, but core identity anchors survive on both.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters are present and clearly recognizable. Aisha retains dark hair, green dress with belt, and similar facial features -- though her hair appears slightly shorter/less voluminous than the reference's shoulder-length wave. Brad keeps the short dark wavy hair, tan jacket, black trousers, green shirt underneath, backpack, and wristwatch. No attribute swapping between characters. Minor drift on Aisha's hair length, otherwise strong continuity.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Disney/Pixar 3D animation, rounded proportions, large expressive eyes, warm color grading, cinematic lighting, CGI render quality (pixar:1.2).. The two characters are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters are present and recognizable. Aisha retains long dark hair, green dress, and necklace -- identity holds well. Brad keeps short dark hair, tan jacket over green shirt, and black trousers -- identity holds. Minor drift: Brad's hair looks slightly less wavy/more styled, and Aisha's face shape is a touch narrower, but no attribute swapping between them.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters are present and recognizable. Aisha retains long dark hair, green dress with belt, and similar facial features. Brad keeps the short dark hair, tan jacket, green shirt underneath, black trousers, and backpack. Minor drift: Aisha's hair appears wavier/longer than the reference's straighter shoulder-length cut, and Brad's jacket looks slightly more structured, but no attribute swapping between characters.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters present. Aisha retains long dark hair, green dress with belt, necklace, and similar face structure. Brad retains short dark hair, tan jacket, and dark trousers. Brad's shirt underneath appears tan/jacket-toned rather than the green shirt visible in reference, and his face reads slightly younger/more open-mouthed, but core anchors hold. No attribute swapping between the two.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters are present. Aisha retains her long black hair, green dress with belt, and necklace -- identity anchors intact. Brad keeps his tan jacket, black trousers, backpack, and short dark hair, though his hair reads slightly shorter/less wavy than the reference. No attribute swapping between the two. Minor drift on Brad's hair texture only.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

nano-banana -- tuned-split

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters present. Aisha retains long black hair, green dress, and similar facial features -- brown belt not visible here as she's seated with book covering waist (minor). Brad retains short dark wavy hair, tan jacket, black trousers, and backpack strap visible. Facial structures match reasonably well. No attribute swapping between the two. Minor drift in Aisha's hair length appearing slightly shorter, but identity holds.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. With them is Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Together they are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters are present. Aisha retains long dark hair, green dress with brown belt (belt visible at waist), gold earrings, and same facial features. Brad retains short dark wavy hair, tan jacket, black trousers, and backpack. Minor drift: Brad's green shirt underneath is less visible and Aisha's dress length/silhouette is similar. No attribute swapping between the two.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. With them is Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Together they are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned-splitref gap
continuity4fidelity5style5cross page consistency4

Both characters present. Aisha retains long black hair, green dress, brown belt, and similar face shape -- strong match. Brad retains short dark hair, tan jacket, black trousers, and backpack; his facial features are close but slightly more stylized/exaggerated. No attribute swapping between the two. Minor drift in Brad's face proportions under the new style.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. With them is Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Together they are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned-splitref gap
continuity4fidelity5style5cross page consistency4

Both characters are present. Brad retains short dark wavy hair, tan jacket, black trousers, and backpack. Aisha keeps long dark hair, green dress, brown belt, and pendant necklace. Minor drift: Aisha gained gold hoop earrings (matches sheet but not reference where she wore small studs), and her hair reads wavier/longer than reference. No attribute swapping between the two.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. With them is Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Together they are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
Graphic Novel

Graphic Novel

Mr Penguin and Brad

Maria Graphic Novel exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptA penguin with wide,curious eyes and clear face and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity3fidelity5style5cross page consistency4

Brad is well preserved: same Latino boy with short dark wavy hair, tan jacket with darker collar, black trousers, similar face. Mr Penguin, however, is missing key anchors visible in the reference: the reference penguin has an orange beak, but the new illustration shows a dark/grey beak with only a hint of orange. Orange webbed feet are not visible (obscured). The penguin's identity anchor of the orange beak is notably wrong, so the weaker character drags the score down.

promptA penguin with wide,curious eyes and clear face and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatimref gap
continuity3fidelity5style5cross page consistency4

Brad is well preserved: same short dark wavy hair, tan jacket with dark collar, gray shirt underneath, black trousers, similar Latino features and facial structure. Mr Penguin, however, drifts significantly -- in the reference the penguin has a BLACK beak (no orange visible) and looks like a realistic Magellanic-style penguin with a white facial band; in the new image the penguin has an ORANGE beak and a very different cartoonier face with huge round eyes and a white eye-mask pattern. The penguin's identity is only loosely recognizable. Weaker character governs, so 3.

promptA penguin with wide,curious eyes and clear face and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity3fidelity4style5cross page consistency4

Brad is well preserved: same short dark wavy hair, tan jacket over cream shirt, black trousers, Latino skin tone -- identity holds. Mr Penguin is weaker: the reference shows a penguin with a pink/orange beak but here the beak is orange-yellow and small, and critically the reference penguin has clear black-and-white markings with wide curious eyes, while the new one looks more cartoonish with a different face shape. No attribute swapping between the two. The weaker penguin match governs the score.

promptA penguin with wide,curious eyes and clear face and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatimref gap
continuity3fidelity5style5cross page consistency4

Brad matches well: same short dark wavy hair, tan jacket with dark collar, black trousers, similar face and skin tone. However, Mr Penguin's identity drifts significantly -- in the reference the penguin has a pinkish/pale beak and pale feet (not the orange specified in the sheet), and appears smaller and more naturalistic. In the new image the penguin lacks the orange beak (beak appears orange-ish but small), is much larger/more stylized, and the proportions are quite different. The penguin is recognizably a penguin but not clearly the same individual as the reference.

promptA penguin with wide,curious eyes and clear face and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tunedref gap
continuity4fidelity5style5cross page consistency4

Both characters are present. Brad retains short dark wavy hair, tan jacket with darker collar, gray/black shirt underneath, and dark trousers -- identity holds strongly. Mr Penguin has the black back, white belly, and wide curious eyes matching the reference; however, the beak in the reference appears dark/black rather than orange, and in the new image it is also dark -- so consistent with reference. Feet are not clearly visible. Minor drift: penguin's face marking (white eye stripe) is slightly more pronounced in the new image but consistent. No attribute swapping between characters.

promptCharacter: the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
beach
beachgpt-image-2 · arm: tunedref gap
continuity4fidelity5style5cross page consistency4

Both characters are present. Brad retains short dark wavy hair, tan jacket with darker collar, layered shirt, and dark trousers -- facial features match closely. Mr Penguin has the black back, white belly, and wide curious eyes, but the beak reads more dark/orange-tipped than fully orange, and the orange webbed feet are not visible in this pose. Minor drift on the penguin's beak color saturation, but identities are unmistakable.

promptCharacter: the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
sledding
sleddinggpt-image-2 · arm: tuned
continuity3fidelity5style5cross page consistency4

Brad is well preserved: short dark hair, tan jacket, cream shirt, black trousers, brown skin tone all carry over. Mr Penguin, however, has lost his orange beak (rendered dark/black instead) and his orange webbed feet are only faintly hinted at -- two key color anchors from the reference are missing. Both characters are present with no attribute swapping, but the penguin's identity anchors have clearly drifted.

promptCharacter: the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
garden
gardengpt-image-2 · arm: tuned
continuity3fidelity5style5cross page consistency4

Brad is well preserved: short dark wavy hair, tan jacket with dark collar, gray shirt, black trousers -- all anchors hold. Mr Penguin is the weaker match: black back and white belly are correct and eyes are visible, but the beak is dark/black rather than the orange shown in the reference. The feet appear orange, partially preserved. Since the beak color is a key reference-visible anchor and it's wrong, continuity drops to 3.

promptCharacter: the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.

gpt-image-2 -- tuned-split

$0.040 / imagecross-page consistency5
library
librarygpt-image-2 · arm: tuned-split
continuity4fidelity5style4cross page consistency5

Both characters present. Brad retains short dark wavy hair, tan jacket, black trousers, and Latino features -- strong match. Mr Penguin has black back, white belly, wide eyes, but his beak appears dark/black rather than the reference's orange beak, and orange feet are not visible (though he's seated). The beak color drift is a minor but notable anchor break. No attribute swapping between characters.

promptCharacter 1: Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. Character 2: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
beach
beachgpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency5

Both characters are present. Brad is essentially unchanged -- same short dark wavy hair, tan jacket, black trousers, similar facial features. Mr Penguin retains the black back, white belly, and wide curious eyes; however, the beak in the new image reads more dark/brown-orange rather than the clearly orange beak in the reference, and the feet appear more muted orange but visible. Minor drift on the penguin's beak color saturation, no attribute swapping between characters.

promptCharacter 1: Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. Character 2: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
sledding
sleddinggpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency5

Both characters present. Brad retains short dark wavy hair, tan jacket, black trousers, Latino skin tone -- strong match. Mr Penguin has black back, white belly, orange beak, orange webbed feet, and curious eyes -- all anchors present. Minor drift: penguin's eyes appear a bit narrower/more stylized than the reference's wide curious look, but identity holds. No attribute swapping between characters.

promptCharacter 1: Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. Character 2: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
garden
gardengpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency5

Both characters are present. Brad retains short dark wavy hair, tan jacket with dark collar, black trousers, and same facial features -- strong match. Mr Penguin has black back, white belly, orange beak, and curious eyes matching the reference; however, the penguin's feet appear less clearly orange/visible here and the face marking pattern is slightly different (more pronounced white eye ring), a minor drift. No attribute swapping between characters.

promptCharacter 1: Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. Character 2: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity3fidelity5style4cross page consistency4

Brad is recognizable: short dark wavy hair, tan jacket, black trousers, backpack strap, similar Latino features -- solid match. Mr Penguin, however, drifts significantly: in the reference he has a realistic African-penguin look with a pinkish/orange beak patch and naturalistic markings, but in the new image he's rendered as a cartoonish round penguin with a bright orange triangular beak and big cartoon eyes. Species/kind is preserved and colors roughly match, but the penguin's proportions and stylization change enough that he reads as a different, more generic penguin. Weaker character governs -- 3.

promptA penguin with wide,curious eyes and clear face and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity3fidelity4style5cross page consistency4

Brad is well preserved: same short dark wavy hair, tan jacket, black trousers, backpack, and Latino features - strong match. Mr Penguin, however, has drifted significantly: in the reference the penguin has a distinctive white face mask with black eye markings and a pink/orange beak, while in the new image the penguin has a mostly black head with white eye patches and a yellow-orange beak, and the feet are not clearly the vivid orange webbed feet. Penguin identity anchors partially lost, so the weaker character governs the score.

promptA penguin with wide,curious eyes and clear face and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity4style4cross page consistency4

Both characters are present. Brad retains short dark wavy hair, tan jacket with darker collar, cream shirt, black trousers, and backpack straps -- identity holds well. Mr Penguin has the black back, white belly, orange beak, and orange feet, with wide curious eyes preserved. Minor drift: penguin appears slightly more stylized/rounder and smaller in proportion, but no attribute swapping between characters.

promptA penguin with wide,curious eyes and clear face and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity3fidelity4style5cross page consistency4

Brad is well preserved: same short dark wavy hair, tan jacket, black trousers, backpack, and similar Latino features -- strong match. Mr Penguin, however, drifted: in the reference the penguin has a clean black head, white belly, and orange beak, but in the new illustration the penguin's face has an odd white/black mask pattern with prominent white eye rings, making it look like a different species. Beak and feet colors roughly hold. The weaker character (penguin) governs the score.

promptA penguin with wide,curious eyes and clear face and a 9th grader (Latino boy:1.1) wearing (tan jacket:1.1) and (black trousers:1.3)in a (botanical garden:1.0). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity3fidelity5style4cross page consistency4

Brad is recognizable: short dark hair, tan jacket, black trousers, similar face structure -- solid match. Mr Penguin, however, has lost his orange beak (rendered dark/pinkish-grey) and his orange webbed feet are now pale peach/pinkish, which are two key reference-visible anchors. Body shape and black/white coloring survive, but the signature orange features are gone, making the penguin a weaker match and dragging the pair score down.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tunedref gap
continuity4fidelity5style4cross page consistency4

Both characters present. Brad retains short dark hair, tan jacket, black trousers, and backpack -- identity holds well. Mr Penguin has black back, white belly, and orange beak, matching the reference. Minor drift: penguin's feet appear more grayish/tan than clearly orange in the new image, and the penguin's proportions are slightly stockier, but no attribute swapping between characters.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity3fidelity5style4cross page consistency4

Both characters are present. Brad retains his short dark hair, tan jacket, and black trousers -- his identity holds well. Mr Penguin, however, has drifted: in the reference he has a distinctive black head with a white facial marking (African penguin look), but in the new illustration his head is white with only a black cap/back, making him look like a more generic cartoon penguin. The orange beak and feet are preserved, but the head coloration -- a key reference-visible anchor -- is wrong.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tunedref gap
continuity3fidelity5style4cross page consistency4

Both characters are present. Brad's anchors hold well: short dark wavy hair, tan jacket, black trousers, backpack, Latino skin tone all preserved. However, Mr Penguin's identity drifted significantly -- in the reference the penguin has no visible orange beak (appears pinkish/small) and the eyes are distinct with a white face patch, but in the new image the penguin has a dark/grey beak rather than orange, and the coloring/markings look more like a gentoo-style penguin with different facial pattern. Orange webbed feet are present. The weaker character (penguin) shows a missing/wrong anchor (beak color), pulling the score down.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same two characters as the attached reference image -- the penguin with black back, white belly, orange beak and orange webbed feet, and the boy with short dark hair, tan jacket, and black trousers, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

nano-banana -- tuned-split

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned-split
continuity4fidelity5style4cross page consistency4

Both characters are present. Brad retains his short dark wavy hair, tan jacket, cream shirt, black trousers, and backpack straps -- identity holds well. Mr Penguin has the black back, white belly, and orange beak/feet from the reference, but his eyes are rendered as small cartoon dots rather than the wide curious eyes seen in the reference, a minor drift. No attribute swapping between characters.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. With them is Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Together they are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned-split
continuity4fidelity5style4cross page consistency4

Both characters are present. Brad matches well: short dark wavy hair, tan jacket, black trousers, Latino complexion -- all preserved. Mr Penguin has the black back, white belly, orange beak, and orange feet, and wide eyes, matching the reference. Minor drift: the penguin's eyes appear larger/more cartoonish with visible white sclera compared to the reference's more subdued eye rendering, and proportions are slightly stylized. No attribute swapping between characters.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. With them is Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Together they are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned-split
continuity4fidelity5style4cross page consistency4

Both characters present. Brad retains short dark wavy hair, tan jacket, black trousers, and Latino complexion - solid match. Mr Penguin has black back, white belly, orange beak, orange webbed feet, and curious eyes - all anchors intact. Minor drift: penguin's proportions are slightly more cartoonish/stylized than reference, and Brad's expression is much more animated (open mouth, wide eyes) but that's scene-appropriate. No attribute swapping between characters.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. With them is Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Together they are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned-splitref gap
continuity3fidelity5style4cross page consistency4

Brad is well preserved: short dark wavy hair, tan jacket, black trousers, backpack straps, and Latino skin tone all match the reference. Mr Penguin, however, drifts significantly -- in the reference the penguin has a small dark beak and a mostly white face with dark eye markings (looking more like a Magellanic/African penguin), while the new illustration gives it a prominent orange beak and a different facial pattern with white eye-rings. The webbed feet are orange in both. Since the penguin's identity anchors partly break (beak color and facial markings changed), the weaker character brings the score down.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. With them is Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Together they are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.

Brad and Aisha

Maria Graphic Novel exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity5fidelity5style5cross page consistency4

Both characters present with anchors intact. Brad: short dark wavy hair, tan jacket with pocket detail, green shirt underneath, black trousers, watch on wrist -- all preserved. Aisha: long black hair, green dress with collar, gold hoop earrings, green bracelet -- all preserved. No attribute swapping between the two. Skin tones and facial features consistent with reference.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Both characters are present and recognizable. Aisha retains long black hair, hoop earrings, green dress, and green bracelet. Brad retains short dark hair, tan jacket, green shirt underneath, black trousers, and dark watch. Minor drift: Aisha's dress appears slightly lighter/brighter green and shows buttons more prominently; Brad's hair looks slightly wavier/messier than the reference's neater style. No attribute swapping between the two.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Both characters are present. Aisha retains long black hair, gold hoop earrings, green dress with brown belt, and green wristband. Brad retains short dark hair, tan jacket over green shirt, black trousers, and dark watch/wristband. Minor drift: Brad's wristband in the reference appears black/watch-like but shows a green band in the new image (possible mild attribute borrowing from Aisha's green wristband), and Aisha's dress collar style is slightly simplified. Overall identity clearly preserved.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Both characters are present and recognizable. Aisha retains long black hair, gold hoop earrings, green collared dress with brown belt, green wristband -- strong match. Brad retains short dark hair, tan jacket, green shirt underneath, black trousers, and dark wristwatch. Minor drift: Brad's hair looks slightly less wavy and Aisha's dress appears shorter in the new image, but core anchors hold. No attribute swapping between the two.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters are present and recognizable. Brad retains short dark wavy hair, tan jacket, black trousers, and green shirt underneath -- all anchors hold. Aisha retains long black hair, gold hoop earrings, and green attire, though her flowing green dress in the reference reads more as a green zip-polo top with green skirt/pants here -- a minor drift in the dress silhouette but color and general identity hold. No attribute swapping between the two.

promptCharacter: the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
beach
beachgpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters are present and clearly identifiable. Brad retains his short dark wavy hair, tan jacket over green shirt, black trousers, and watch. Aisha retains her long black hair, green dress with collar/zip, brown belt, hoop earrings, and green wristband. No attribute swapping detected. Minor drift: Aisha's dress appears shorter/knee-length in the beach scene vs. long in reference, but this is a minor anchor change.

promptCharacter: the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
sledding
sleddinggpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Both characters present. Aisha retains long black hair, gold hoop earrings, green dress with zipper placket and brown belt, and green wristband -- strong match. Brad retains short dark hair, tan jacket, green shirt underneath, black trousers, and watch -- strong match. Minor drift: Brad's jacket appears slightly shorter/different cut and Aisha's dress hem behavior differs, but no attribute swapping between them. Identity clearly preserved.

promptCharacter: the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
garden
gardengpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

Both characters present and clearly recognizable. Brad retains short dark wavy hair, tan jacket with green cuffs, green shirt underneath, black trousers, and wristwatch. Aisha retains long black hair, gold hoop earrings, green collared dress with zipper, brown belt, and green wristband. No attribute swapping between the two.

promptCharacter: the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.

gpt-image-2 -- tuned-split

$0.040 / imagecross-page consistency5
library
librarygpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency5

Both characters present and recognizable. Brad retains short dark wavy hair, tan jacket over green shirt, black trousers -- solid match. Aisha retains long black hair, gold hoop earrings, green dress with brown belt, green wristband -- solid match. Minor drift: Aisha's dress in the new image appears slightly lighter/more lime green and her collar/zipper detail is a bit softened, but no attribute swapping between characters.

promptCharacter 1: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Character 2: Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
beach
beachgpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency5

Both characters present. Aisha retains long black hair, green dress, brown belt, gold hoop earrings, and green wristband -- all anchors intact. Brad retains short dark wavy hair, tan jacket with green cuffs, green shirt underneath, black trousers, and dark watch. Minor drift: Brad's watch appears larger/more prominent and he's barefoot (acceptable for beach). No attribute swapping between characters. Faces are recognizably the same.

promptCharacter 1: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Character 2: Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
sledding
sleddinggpt-image-2 · arm: tuned-split
continuity4fidelity5style5cross page consistency5

Both characters are present and recognizable. Brad retains short dark wavy hair, tan jacket, green shirt underneath, black trousers, and wrist watch. Aisha retains long black hair, gold hoop earrings, green outfit with brown belt, and green wristband. Minor drift: Aisha's dress in the new image looks more like a green jumpsuit/pants outfit rather than the long dress in the reference, likely due to the sledding pose but still a small anchor drift. No attribute swapping between characters.

promptCharacter 1: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Character 2: Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
garden
gardengpt-image-2 · arm: tuned-split
continuity5fidelity5style5cross page consistency5

Both characters are present and clearly the same. Brad retains short dark wavy hair, tan jacket with green cuffs, green tee, black trousers, and dark watch. Aisha retains long black hair, gold hoop earrings, green collared dress with zipper placket, brown belt, and green wristband. No attribute swapping between them.

promptCharacter 1: Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. Character 2: Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Keep each character's features strictly their own -- no attribute may move from one character to the other. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity4style4cross page consistency4

Both characters are present and identifiable. Aisha retains long black hair, hoop earrings, and green dress with collar/zipper detail. Brad retains short dark hair, tan jacket, green shirt underneath, black trousers, and watch. Minor drift: Aisha's dress collar looks slightly different and Brad's hair appears less wavy/more flat, but overall identity holds for both.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Both characters are present. Brad retains short dark wavy hair, tan jacket, green shirt underneath, black trousers, and watch -- all anchors intact. Aisha retains long black hair, brown skin, and a green dress, though the dress style shifted from a collared button-front midi with belt to a sleeveless sundress with thin straps. Her hoop earrings from the reference are not visible. No attribute swapping between characters. Minor drift on Aisha's outfit style, but identity clearly holds.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Both characters are present and recognizable. Aisha retains long black hair, gold hoop earrings, green collared dress with zipper, brown belt, and green wristband. Brad retains short dark hair, tan jacket with green cuffs, green shirt underneath, and black trousers. Minor drift: Brad appears slightly younger/softer-featured and is now smiling broadly, and his jacket sleeves look pushed up, but no attribute swapping between the two is evident.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Both characters present. Aisha: long black hair, green dress with collar and brown belt, gold hoop earrings -- all preserved. Brad: short dark hair, tan jacket with green cuffs, green shirt underneath, black trousers, wristwatch -- all preserved. Brad's hair looks slightly straighter/less wavy than the reference, a minor drift. No attribute swapping between characters.

promptIn a (botanical garden:0.5) a 9th grader Latino boy wearing (tan jacket:1.1) and (black trousers:1.3) and a 9th grade Asian Indian girl (Indian girl:1.4) wearing (green dress:1.3). Bold-ink comic book/graphic novel style, bold ink outlines, cel-shaded (graphic novel:1.2). The two characters are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Both characters are present and clearly recognizable. Aisha retains long black hair, hoop earrings, green dress with zipper detail, and green wristband. Brad keeps short dark hair, tan jacket with sleeve patch, green shirt underneath, black trousers, and dark watch. Minor drift: Brad's hair reads slightly straighter/less wavy, and Aisha's expression is softer, but no attribute swapping between the two.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity3fidelity5style4cross page consistency4

Both characters are present. Aisha retains her long black hair and green dress -- though the dress style changed from a collared long dress to a sleeveless sundress, and her hoop earrings are gone. Brad has lost his defining tan jacket and green shirt -- he now wears a plain tan t-shirt and dark shorts instead of black trousers. His short dark hair matches, but the signature jacket (a key anchor) is missing, dropping recognizability.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity4fidelity5style3cross page consistency4

Both characters are present and recognizable. Aisha retains long black hair, green short-sleeve collared dress with brown belt, and gold hoop earrings. Brad retains short dark hair, tan jacket with green cuffs, green shirt underneath, and black trousers. Minor drift: Aisha's dress collar/zipper detail is less defined and Brad's hair looks slightly less wavy, but no attribute swapping between them.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity4fidelity4style5cross page consistency4

Both characters are present and clearly recognizable. Brad retains short dark wavy hair, tan jacket with green undershirt, black trousers, and wristwatch. Aisha keeps her long black hair, gold hoop earrings, green top with zipper detail, and brown belt. Minor drift: Aisha's full-length green dress in the reference has become green pants/jumpsuit in the new image -- a notable clothing anchor change, though the green color and belt survive. No attribute swapping between the two.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same two characters as the attached reference image -- the boy with short dark hair, tan jacket, and black trousers, and the girl with long black hair and a green dress, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

nano-banana -- tuned-split

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned-split
continuity5fidelity5style4cross page consistency4

Both characters present and faithfully reproduced. Brad: short dark wavy hair, tan jacket with green tee visible, black trousers, watch on wrist -- all match. Aisha: long black hair, gold hoop earrings, green collared dress with zip detail -- all match. No attribute swapping between characters.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. With them is Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Together they are sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned-split
continuity4fidelity5style5cross page consistency4

Both characters are present and recognizable. Brad retains short dark wavy hair, tan jacket with green shirt underneath, and black trousers -- strong match. Aisha retains long black hair, gold hoop earrings, and green outfit with brown belt, but her dress appears to have become more pants-like/jumpsuit in the sitting pose, a minor drift. No attribute swapping between characters.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. With them is Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Together they are building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned-split
continuity4fidelity5style4cross page consistency4

Both characters are present. Aisha retains long black hair, gold hoop earrings, green dress with brown belt, and green wristband -- solid match. Brad retains short dark hair, tan jacket, black trousers, and green undershirt. Minor drift: Brad's hair looks slightly less wavy and his face reads a touch younger/rounder, but identity holds. No attribute swapping between characters.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. With them is Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Together they are riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned-split
continuity4fidelity4style5cross page consistency4

Both characters are present and recognizable. Aisha retains long black hair, gold hoop earrings, green dress with brown belt -- all anchors hold. Brad retains short dark hair, tan jacket, green shirt underneath, black trousers, and watch. Minor drift: Brad's hair looks slightly less wavy and more slicked, and Aisha's green wristband from the reference is not visible, but no attribute swapping between characters.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows Brad -- the same boy as in the attached reference image: short dark wavy hair, tan jacket, black trousers. With them is Aisha -- the same girl as in the attached reference image: long black hair, gold hoop earrings, green dress with a brown belt. Together they are planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for each character's identity and the target art style, keeping every feature on its own character and every surface free of lettering, captions, and panel borders.

Full written results

the complete documents behind this page -- method, every table, the reads, and the standing caveats

Last updated: 2026-07-31

Experiment 6 -- multi-character pairs: results

Experiment 3's protocol with two characters per image. Each pair reference is a single SIC mint showing both characters; the judge runs in pair mode -- it checks each character's anchors separately, the weaker character's identity match governs the continuity score, and it is explicitly instructed to flag attribute swapping or blending between the two. Two pairs (Mr Penguin with Brad, Brad with Aisha), two styles so far (3D Animation, Graphic Novel), lead model pair only, both prompt arms, 4 scenes per arm. 64 renders, all judged inline by opus, zero render failures. Cost: ~$2.53 ($0.040/image gpt-image-2, $0.039/image nano-banana, reference input included).

Results: Mr Penguin + Brad

StyleModelA: sic-verbatim (cont/style)B: tuned (cont/style)Cont delta
3d-animationgpt-image-24.00 / 5.004.00 / 5.000.00
3d-animationnano-banana4.00 / 5.003.50 / 5.00-0.50
graphic-novelgpt-image-23.00 / 5.003.50 / 5.00+0.50
graphic-novelnano-banana3.25 / 4.503.25 / 4.000.00

Results: Brad + Aisha

StyleModelA: sic-verbatim (cont/style)B: tuned (cont/style)Cont delta
3d-animationgpt-image-24.25 / 5.004.00 / 5.00-0.25
3d-animationnano-banana4.00 / 5.004.00 / 5.000.00
graphic-novelgpt-image-24.25 / 5.004.25 / 5.000.00
graphic-novelnano-banana4.00 / 4.753.75 / 4.00-0.25

Cross-page consistency is 4 on every arm of both pairs.

The read

  1. A second character costs continuity only when one member is weak. Brad + Aisha -- two humans, both anchor-rich -- holds 3.75-4.25, on par with Brad's solo experiments. Mr Penguin + Brad drops to 3.00-3.50 on graphic-novel, and the judge's notes put the loss almost entirely on the penguin (his orange beak keeps rendering dark, the same design-ambiguity his solo graphic-novel arms showed in experiment 3b) while Brad's anchors hold in the same frames. Pair continuity behaves like the minimum of the members' solo continuity, which is exactly what weaker-character-governs scoring should produce. The production implication: multi-character pages don't need a new model or new prompting -- they need every character to pass reference QC individually before sharing a page.
  2. No attribute swapping anywhere. The judge checked each of the 64 renders for features migrating between characters and never flagged a swap -- no penguin colors on Brad, no jacket on the penguin, nothing between Brad and Aisha. The failure mode the round was designed to catch did not occur on either model.
  3. Both characters showed up in 63 of 64 renders. The one real presence failure: nano-banana's tuned 3d sledding scene rendered Mr Penguin squished against Brad's back like an accessory rather than a character. That arm's -0.50 is mostly this one scene.
  4. Prompt tuning stops paying in pair mode. Deltas run -0.50 to +0.50 and average slightly negative -- the first round where tuned doesn't win. The tuned arm's identity lock is one shared clause covering both characters, so per-character identity language is diluted exactly where it has two identities to carry. Worth a follow-up arm with separate per-character locks before concluding tuning has nothing to offer pairs.
  5. gpt-image-2 meets or beats nano-banana in every matched cell, and style adherence stays high for both (gpt-image-2 hit 5.00 style on 8 of 8 arms; nano dipped to 4.00-4.75 on graphic-novel). The lead-pair ordering from experiments 3/3b survives multi-character unchanged.

Standing caveats: n=4 per cell; two pairs, two styles; the penguin-brad graphic-novel reference itself hides some sheet anchors (reference_gap flagged on those arms), so his floor there is partly a reference problem -- consistent with, not separate from, the experiment 3b reference-strength finding.

Remaining for this experiment: vibrant-fantasy and fantasy-art pair exemplars are drafted in the SIC log but commented out -- those cells run once the exemplars are minted and frozen.

_Last updated: 2026-08-05_

Experiment 6b -- per-character identity locks for multi-character pages

The question

Experiment 6 found that prompt tuning, which wins everywhere else in the program, stops paying once two characters share a page: deltas ran -0.50 to +0.50 and averaged slightly negative. The suspected cause was dilution. The tuned arm carries identity in a single shared clause -- "the same two characters as the attached reference image -- the penguin with black back... and the boy with short dark hair..." -- so one sentence has to hold two identities at once.

This experiment splits that clause. Each character gets its own labeled lock, plus one explicit no-blend instruction. Everything else is held fixed: same pairs, same styles, same references, same scenes, same judge, same pair-mode rubric.

The arms

  • A -- sic-verbatim (from E6): the production SIC prompt unmodified.
  • B -- tuned, shared lock (from E6): the per-model tuned idiom with one combined identity clause.
  • C -- tuned-split (new): the same tuned idiom, but with one lock per character.
    • gpt family: Character 1: Mr Penguin -- the same penguin as in the attached reference image: black back, white belly, orange beak, orange webbed feet, wide curious eyes. Character 2: Brad -- ... Keep each character's features strictly their own -- no attribute may move from one character to the other.
    • Gemini: each character introduced in its own sentence (It shows &lt;lock 1&gt;. With them is &lt;lock 2&gt;.), closing with "keeping every feature on its own character."

2 pairs x 2 styles x 2 models x 4 scenes = 32 renders, opus-judged in pair mode.

Results: character continuity by arm

PairStyleModelA: sicB: shared lockC: split locksC vs BC vs A
Brad + Aisha3d-animationgpt-image-24.254.004.000.00-0.25
Brad + Aisha3d-animationnano-banana4.004.004.000.000.00
Brad + Aishagraphic-novelgpt-image-24.254.254.250.000.00
Brad + Aishagraphic-novelnano-banana4.003.754.25+0.50+0.25
Mr Penguin + Brad3d-animationgpt-image-24.004.004.000.000.00
Mr Penguin + Brad3d-animationnano-banana4.003.504.00+0.500.00
Mr Penguin + Bradgraphic-novelgpt-image-23.003.504.00+0.50+1.00
Mr Penguin + Bradgraphic-novelnano-banana3.253.253.75+0.50+0.50

Mean C-vs-B delta +0.25. Four cells improved by half a point, four unchanged, none regressed.

The read

Split locks recover the payoff that pair mode had taken away, and the recovery is not spread evenly -- it lands exactly where dilution was doing damage. Every cell that gained was a cell where the shared lock had failed to beat its own control. Brad + Aisha on 3d-animation and graphic-novel with gpt-image-2 were already at 4.00-4.25 with nothing left to recover, and they did not move. Mr Penguin + Brad on graphic-novel -- the worst cell in all of experiment 6, where the penguin's own reference is design-ambiguous and the shared clause had to carry him alongside a well-anchored human -- gained a full point over control and half a point over the shared lock.

That is the shape a dilution effect should have. If the gain had appeared uniformly, the more likely explanation would be that the split prompt is simply longer or more emphatic; instead it appears only where one identity was being crowded out by the other.

The no-swap property survives. All 32 renders were checked for attribute migration between characters, and the judge asserted a swap in none of them -- the same result experiment 6 got with the shared lock. Splitting the identity language did not introduce the failure mode that separating characters might plausibly cause. Both characters were present in 31 of 32 renders.

Cross-page consistency held or improved: 4 on six arms, 5 on two (both gpt-image-2 graphic-novel), versus a flat 4 across every arm in experiment 6.

Cost: 32 renders, ~$1.26 all-in.

What this changes for production

Multi-character pages should compose identity per character, not per page. Concretely, the prompt assembler should emit one labeled identity clause per character on the page plus a single no-blend sentence, rather than concatenating the cast into one description. This costs nothing at inference time -- it is the same prompt budget rearranged -- and it is worth up to a full point of continuity on the hardest pairs.

The result also narrows what still needs reference-QC attention: the shared lock was masking a weak character, so a page with one strong and one weak reference now degrades more gracefully. That does not remove the reference-QC gate, but it lowers the penalty for the case the gate is most likely to miss.

Incidents

One render (Brad + Aisha, 3d-animation, gpt-image-2, beach scene) was rejected by OpenAI's safety system on the first attempt and rendered clean on an immediate retry with the byte-identical prompt. This is the program's second OpenAI benign block and the first that proved transient -- the earlier "Brad waters a plant" block reproduced three times across two runs and only cleared when the prompt was restructured. The two specimens together define the fallback ladder: retry once (catches transients, ~$0.03), then restructure (catches deterministic phrasing blocks), then ship the page imageless as the last resort. The patched render carries a note in its record.

Standing caveats

n=4 per cell; two pairs; two styles. Direction is consistent across all eight cells (no regressions), which is stronger evidence than any single cell's magnitude. Trios are untested here -- dilution should get worse with three characters, and that cell can run against the three existing per-character references without minting a group exemplar.

Story Studio setting continuity -- experiment 4

Findings
  • Repeating the full room description on every page is what works. A setting sheet lists the room's five checkable details -- round window, red beanbag, oak bookshelf, striped rug, chalkboard easel. Repeated word-for-word on every page, it brought back 95% of those details on nano-banana and 85% on gpt-image-2. Name the room without describing it and you get 5-15%. That gap is far too big to be noise, which is why this one is stated flatly.
  • Shortening the description does not work. We expected naming the top three details to buy most of the benefit at a fraction of the prompt length. It does not: the model keeps the three you name and quietly drops the other two, landing at ~40% recall. At least for this room, how much detail you give is what matters.
  • Reusing one random seed scores perfectly and is still unusable. Stable Image Core is the only model here that exposes a seed. Holding it constant across all four pages produces a top consistency score -- but only because the pages come out as near-duplicates of each other, so the room never gets re-framed and the night page never actually goes dark. It games the metric rather than solving the problem.
  • gpt-image-2 never volunteered lettering; stable-image-core did on every cell, nano-banana on two arms. Caveat added 2026-08-04: every experiment 4 arm carries the blank-surface clause, so this is a result about the clause working, not about gpt-image-2 being text-shy. Experiment 5 removed the clause and gpt-image-2 lettered 14 of 15 cells.
  • stable-image-core is not a candidate to render pages -- it is here as production's baseline and as the model that mints our style exemplars, which it does better than anything else measured. Experiment 4 tests the half of the job it fails: 15% recall at its best prompt-only cell, no graphic-novel style adherence, and lettering in every cell. Read its rows as the size of the problem.
Recommended next steps
  • Adopt full setting sheets in production prompt assembly -- the win is pure prompt text, no pipeline change.
  • Run the remaining two settings plus a fresh-settings replication set before locking the recipe.
  • Price any future setting-reference feature against these prompt-only numbers.
Read the full written results for this experiment ↓method, every table, the reads, and the standing caveats

Can prompts alone keep four story pages in the same room? No reference images on any model -- this is production's floor.

  • The material: one 4-page story (Maria and Mr Penguin in the reading corner, characters by text only); pages recompose the room -- different areas, wider framing, and a night version.
  • Arm A -- name-only: the room appears only as its name. The control floor.
  • Arm B -- full setting sheet: all five room anchors, verbatim on every page.
  • Arm C -- compact lock: just the top three anchors.
  • Arm D -- seed-hold (Stable Image Core only): arm B plus the same seed on all four pages.
  • Scoring: per-card anchor checklists; header chips carry cross-page room consistency, night geometry vs palette, pairwise same-room verdicts, and the unprompted-text check. Style fixed to graphic-novel.
setting the reading cornerjudge claude-opus-4-7 via Bedrockarms name-only / sheet / lock / seed-holdruns 2026-08-03

stable-image-core

stable-image-core -- A -- name-only (control)

$0.04 / imagecross-page consistency3night geometry3night palette4same-room pairs1/3unprompted text4

Round window never appears -- it is always a rectangular multi-pane window. Red beanbag chair is missing on every page (replaced by window bench, orange cushion, gray cushion, or bare floor). Brass globe never appears atop the bookshelf; page 2 substitutes a pumpkin. Striped blue-and-yellow rug never appears; page 2 shows an orange cushion/rug, others show bare wood. Chalkboard easel with sun is entirely absent on all pages. Bookshelf configuration drifts: single left-side shelf (p1) → tall corner shelf with framed pictures (p2) → symmetric flanking shelves (p3, p4). Windowsill contents change between potted plant (p1), candle+flower vase (p2), stray book (p3), two small potted plants+pencil cup (p4).

page1-read
page1-readstable-image-core
  • △ a round window with rain outside rectangular multi-pane window, not round; rain present
  • ✗ a red beanbag chair child sits on wooden window bench
  • △ a tall oak bookshelf with a brass globe on top bookshelf present at left but no brass globe
  • ✗ a striped blue-and-yellow rug no rug visible
  • ✗ a chalkboard easel with a sun drawn on it no chalkboard shown
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page2-search
page2-searchstable-image-core
  • △ a round window with rain outside rectangular multi-pane windows, rain visible
  • ✗ a red beanbag chair orange cushion/floor seating instead
  • △ a tall oak bookshelf with a brass globe on top tall bookshelf present but topped with a pumpkin, not a globe
  • △ a striped blue-and-yellow rug orange cushion/rug, no blue-yellow stripes
  • ✗ a chalkboard easel with a sun drawn on it no chalkboard
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, kneels at the foot of the tall bookshelf searching the lower shelves for a lost little brass key, shown from a low angle that emphasizes the bookshelf. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page3-talk
page3-talkstable-image-core
  • △ a round window with rain outside rectangular multi-pane window, rain present
  • ✗ a red beanbag chair gray cushion instead
  • △ a tall oak bookshelf with a brass globe on top bookshelves flank the window but no globe on top
  • ✗ a striped blue-and-yellow rug bare wood floor
  • ✗ a chalkboard easel with a sun drawn on it no chalkboard
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, stands talking with Mr Penguin, a penguin with a black back, white belly, an orange beak, orange webbed feet, and wide curious eyes, the two of them together on the rug beside the chalkboard easel, shown in a wide shot that takes in the whole corner. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page4-night
page4-nightstable-image-core
  • △ a round window with rain outside rectangular multi-pane window, rain outside at night
  • ✗ a red beanbag chair child sits cross-legged on floor
  • △ a tall oak bookshelf with a brass globe on top tall bookshelves present, no globe
  • ✗ a striped blue-and-yellow rug bare wood floor
  • ✗ a chalkboard easel with a sun drawn on it no chalkboard; a wall lamp instead
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

stable-image-core -- B -- full setting sheet

$0.04 / imagecross-page consistency2night geometry2night palette1same-room pairs0/3unprompted text3

Rug: changes from concentric bullseye (p1-3) to plain blue with yellow dot (p4); never a blue-and-yellow stripe. Chair: yellow tub chair on p1, red beanbag p2-4 (position moves). Chalkboard easel: freestanding on p1, mounted/relocated on p2, on wall on p3, absent on p4 (replaced by teepee). Globe: absent on p1, on shelf p2, on floor p3, on side table p4 -- never on top of the bookshelf as specified. Sun: painted on wall p1, replaced by yellow ceiling ball p2, yellow disk on wall p3, sun on wall p4. Window: arched throughout, never round; rain never shown. Ceiling fixture changes each page (white globe, yellow orb, small dome, yellow drum lamp).

page1-read
page1-readstable-image-core
  • △ a round window with rain outside arched window, no rain visible
  • △ a red beanbag chair yellow round chair present; red ball on floor instead
  • △ a tall oak bookshelf with a brass globe on top tall bookshelf present, but no brass globe on top
  • △ a striped blue-and-yellow rug round rug with concentric blue/yellow/red bands, not stripes
  • △ a chalkboard easel with a sun drawn on it chalkboard easel present with writing; sun is painted on wall, not on chalkboard
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page2-search
page2-searchstable-image-core
  • △ a round window with rain outside arched window with trees, no rain
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelf present, globe sits on a shelf not on top
  • △ a striped blue-and-yellow rug round concentric-ring rug, not stripes
  • △ a chalkboard easel with a sun drawn on it chalkboard present with words, no sun; sun is a yellow ceiling ball
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, kneels at the foot of the tall bookshelf searching the lower shelves for a lost little brass key, shown from a low angle that emphasizes the bookshelf. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page3-talk
page3-talkstable-image-core
  • △ a round window with rain outside arched window with trees, no rain
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top two bookshelves; a small globe stands separately on the floor
  • △ a striped blue-and-yellow rug concentric ring rug
  • △ a chalkboard easel with a sun drawn on it chalkboard on wall with faint circles, no clear sun; separate yellow disk on wall
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, stands talking with Mr Penguin, a penguin with a black back, white belly, an orange beak, orange webbed feet, and wide curious eyes, the two of them together on the rug beside the chalkboard easel, shown in a wide shot that takes in the whole corner. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page4-night
page4-nightstable-image-core
  • △ a round window with rain outside arched window, clear daylight, rooftops visible, no rain
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top tall bookshelves flank; globe sits on a small side table
  • △ a striped blue-and-yellow rug plain round blue rug with yellow center, no stripes
  • ✗ a chalkboard easel with a sun drawn on it no chalkboard visible; a yellow teepee stands in its place; sun is on wall
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

stable-image-core -- C -- compact 3-anchor lock

$0.04 / imagecross-page consistency2night geometry1night palette5same-room pairs0/3unprompted text3

Window changes shape/count every page (arched single → double rectangular → single rectangular → curved bay at night). The striped blue-and-yellow rug never appears; instead a gray irregular rug (p1), round gray rug (p2), gray rectangular rug (p3), and small beige round rug (p4) are shown. The brass globe on top of the oak bookshelf never appears. The chalkboard easel with a sun is absent on all pages. Red seating drifts from beanbag (p1-3) to armchair + pouf (p4). Wall art rearranges each page. Ceiling changes from implied flat (p1) to visible tiled with pendant (p2) to none (p3) to red pendant lamp (p4). Small side bookshelf appears only in p3 and p4.

page1-read
page1-readstable-image-core
  • △ a round window with rain outside rectangular window with arched top, not round
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top tall bookshelves present but no brass globe visible
  • △ a striped blue-and-yellow rug gray irregular rug, not striped blue/yellow
  • ✗ a chalkboard easel with a sun drawn on it no chalkboard visible
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page2-search
page2-searchstable-image-core
  • △ a round window with rain outside two large rectangular grid windows
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelves present, no brass globe
  • △ a striped blue-and-yellow rug round gray rug
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, kneels at the foot of the tall bookshelf searching the lower shelves for a lost little brass key, shown from a low angle that emphasizes the bookshelf. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page3-talk
page3-talkstable-image-core
  • △ a round window with rain outside large rectangular grid window
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top shelves present, no globe
  • △ a striped blue-and-yellow rug plain gray rug
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, stands talking with Mr Penguin, a penguin with a black back, white belly, an orange beak, orange webbed feet, and wide curious eyes, the two of them together on the rug beside the chalkboard easel, shown in a wide shot that takes in the whole corner. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page4-night
page4-nightstable-image-core
  • △ a round window with rain outside curved bay windows at night with moon, rain visible, not round
  • △ a red beanbag chair red armchair and red pouf instead of beanbag
  • △ a tall oak bookshelf with a brass globe on top tall oak shelves, no brass globe
  • △ a striped blue-and-yellow rug small round beige rug
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

stable-image-core -- D -- seed-hold (sheet + same seed, SIC only)

$0.04 / imagecross-page consistency5night geometry5night palette1same-room pairs3/3unprompted text4

Across all four pages the images are essentially duplicates, so nothing drifts between pages - but multiple anchors are wrong relative to the setting sheet: the window is arched rather than round and shows no rain; the bookshelf has no brass globe on top (a yellow paper lantern hangs from the ceiling instead); the rug is a circular multicolor bullseye rather than a blue-and-yellow striped rug; the chalkboard is a small wall-mounted board with gibberish writing rather than an easel with a sun drawn on it; and the designated night page (page 4) is not depicted as night.

page1-read
page1-readstable-image-core
  • △ a round window with rain outside arched window (not round), no visible rain, autumn trees
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top tall oak bookshelf present but no brass globe on top
  • △ a striped blue-and-yellow rug round rug with blue/yellow/red concentric shapes, not striped
  • △ a chalkboard easel with a sun drawn on it small wall-mounted chalkboard with lettering, not an easel, no sun
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page2-search
page2-searchstable-image-core
  • △ a round window with rain outside same arched window, no rain
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top no globe on top
  • △ a striped blue-and-yellow rug same circular multicolor rug
  • △ a chalkboard easel with a sun drawn on it wall chalkboard with text, not easel/sun
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, kneels at the foot of the tall bookshelf searching the lower shelves for a lost little brass key, shown from a low angle that emphasizes the bookshelf. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page3-talk
page3-talkstable-image-core
  • △ a round window with rain outside arched window, no rain
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top no globe
  • △ a striped blue-and-yellow rug same circular rug, not striped
  • △ a chalkboard easel with a sun drawn on it wall chalkboard with text
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, stands talking with Mr Penguin, a penguin with a black back, white belly, an orange beak, orange webbed feet, and wide curious eyes, the two of them together on the rug beside the chalkboard easel, shown in a wide shot that takes in the whole corner. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page4-night
page4-nightstable-image-core
  • △ a round window with rain outside arched window, daylight visible - NOT nighttime
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top no globe
  • △ a striped blue-and-yellow rug same circular rug
  • △ a chalkboard easel with a sun drawn on it wall chalkboard with text
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

gpt-image-2

gpt-image-2 -- A -- name-only (control)

$0.032 / imagecross-page consistency2night geometry2night palette5same-room pairs0/3unprompted text0

Window shape drifts: rectangular in p1-3, round only in p4. Beanbag color/presence drifts: orange-red p1, absent p2, purple p3, red p4. Rug drifts entirely: dark blue solid p1, red patterned p2, round teal p3, barely visible p4 -- striped blue-and-yellow never appears. Brass globe on bookshelf never appears on any page. Chalkboard easel appears only p3 and lacks the drawn sun. Bookshelf position moves around the room between pages.

page1-read
page1-readgpt-image-2
  • △ a round window with rain outside window is rectangular/grid, not round
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelf present but no brass globe on top
  • △ a striped blue-and-yellow rug rug is dark blue, no yellow stripes
  • ✗ a chalkboard easel with a sun drawn on it no easel visible
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page2-search
page2-searchgpt-image-2
  • △ a round window with rain outside rectangular window with rain
  • ✗ a red beanbag chair armchair instead, no beanbag
  • △ a tall oak bookshelf with a brass globe on top large bookshelf, no globe
  • △ a striped blue-and-yellow rug red patterned rug
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, kneels at the foot of the tall bookshelf searching the lower shelves for a lost little brass key, shown from a low angle that emphasizes the bookshelf. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page3-talk
page3-talkgpt-image-2
  • △ a round window with rain outside rectangular window with rain
  • △ a red beanbag chair beanbag is purple, not red
  • △ a tall oak bookshelf with a brass globe on top bookshelf present, plant on top instead of brass globe
  • △ a striped blue-and-yellow rug round teal rug
  • △ a chalkboard easel with a sun drawn on it easel present but blank, no sun drawn
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, stands talking with Mr Penguin, a penguin with a black back, white belly, an orange beak, orange webbed feet, and wide curious eyes, the two of them together on the rug beside the chalkboard easel, shown in a wide shot that takes in the whole corner. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page4-night
page4-nightgpt-image-2
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelf visible, no brass globe
  • ✗ a striped blue-and-yellow rug rug barely visible, no stripes
  • ✗ a chalkboard easel with a sun drawn on it no easel
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

gpt-image-2 -- B -- full setting sheet

$0.032 / imagecross-page consistency3night geometry4night palette5same-room pairs3/3unprompted text0

Beanbag moves from left-of-window (p1) to under-window (p2) to left-of-bookshelf (p3,p4). Chalkboard easel moves from far right (p1) to center-left beside beanbag (p2) to right of bookshelf (p3,p4). Brass gold globe on page 1 becomes a blue-and-green Earth globe on pages 2-4. Bookshelf proportions shrink on page 3. A small table lamp appears only on page 4.

page1-read
page1-readgpt-image-2
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • ✓ a tall oak bookshelf with a brass globe on top
  • ✓ a striped blue-and-yellow rug
  • ✓ a chalkboard easel with a sun drawn on it
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page2-search
page2-searchgpt-image-2
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top globe now blue/green not brass
  • ✓ a striped blue-and-yellow rug
  • ✓ a chalkboard easel with a sun drawn on it
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, kneels at the foot of the tall bookshelf searching the lower shelves for a lost little brass key, shown from a low angle that emphasizes the bookshelf. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page3-talk
page3-talkgpt-image-2
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelf shorter; globe is blue/green
  • ✓ a striped blue-and-yellow rug
  • ✓ a chalkboard easel with a sun drawn on it
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, stands talking with Mr Penguin, a penguin with a black back, white belly, an orange beak, orange webbed feet, and wide curious eyes, the two of them together on the rug beside the chalkboard easel, shown in a wide shot that takes in the whole corner. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page4-night
page4-nightgpt-image-2
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top globe is blue/green, not brass
  • ✓ a striped blue-and-yellow rug
  • ✓ a chalkboard easel with a sun drawn on it
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

gpt-image-2 -- C -- compact 3-anchor lock

$0.032 / imagecross-page consistency3night geometry4night palette5same-room pairs3/3unprompted text0

Window frame changes from a frameless round disc (p1) to a cross-muntin round window (p2-4). Rug is absent on p1 and p4, appears as a red patterned oriental rug on p2, and as a round red/blue patterned rug on p3 -- never the specified blue-and-yellow stripes. Brass globe only appears on p3 and is missing from p1, p2, p4. Chalkboard easel only appears on p3, and the sun drawing on it is absent (only faint smudges). Bookshelf position stays right side throughout. Beanbag stays red and roughly center/left.

page1-read
page1-readgpt-image-2
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelf present, no globe visible
  • ✗ a striped blue-and-yellow rug bare wood floor
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page2-search
page2-searchgpt-image-2
  • △ a round window with rain outside now has cross muntins, previously frameless
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelf yes, no globe
  • △ a striped blue-and-yellow rug rug is red/patterned, not blue-yellow striped
  • ✗ a chalkboard easel with a sun drawn on it not shown
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, kneels at the foot of the tall bookshelf searching the lower shelves for a lost little brass key, shown from a low angle that emphasizes the bookshelf. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page3-talk
page3-talkgpt-image-2
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • ✓ a tall oak bookshelf with a brass globe on top
  • △ a striped blue-and-yellow rug round red/blue patterned rug, not striped blue-yellow
  • △ a chalkboard easel with a sun drawn on it easel present but no sun drawing, just faint marks
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, stands talking with Mr Penguin, a penguin with a black back, white belly, an orange beak, orange webbed feet, and wide curious eyes, the two of them together on the rug beside the chalkboard easel, shown in a wide shot that takes in the whole corner. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page4-night
page4-nightgpt-image-2
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelf visible, globe not shown
  • ✗ a striped blue-and-yellow rug wood floor only
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

nano-banana

nano-banana -- A -- name-only (control)

$0.039 / imagecross-page consistency2night geometry2night palette5same-room pairs0/3unprompted text1

Window: rectangular tall (p1) -> rectangular sash (p2) -> tall multi-pane wall (p3) -> round porthole (p4); only p4 matches the 'round window' anchor. Beanbag: blue on p1, absent p2-3, blue again p4; never the specified red. Bookshelf: present all pages but never has the brass globe on top. Rug: absent p1, patterned rectangular p2, circular concentric grey p3, absent p4; never the striped blue-and-yellow specified. Chalkboard easel: only appears on p3 and is blank (no sun); absent on p1, p2, p4.

page1-read
page1-readnano-banana
  • △ a round window with rain outside rectangular window, not round
  • △ a red beanbag chair blue beanbag instead of red
  • △ a tall oak bookshelf with a brass globe on top bookshelf present but no brass globe on top
  • ✗ a striped blue-and-yellow rug no rug visible, plain floor
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page2-search
page2-searchnano-banana
  • △ a round window with rain outside rectangular window with rain, not round
  • ✗ a red beanbag chair no beanbag visible
  • △ a tall oak bookshelf with a brass globe on top tall bookshelf present, no globe on top
  • △ a striped blue-and-yellow rug a patterned rug is on the floor but not striped blue-and-yellow
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, kneels at the foot of the tall bookshelf searching the lower shelves for a lost little brass key, shown from a low angle that emphasizes the bookshelf. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page3-talk
page3-talknano-banana
  • △ a round window with rain outside large rectangular multi-pane window with rain
  • ✗ a red beanbag chair not visible
  • △ a tall oak bookshelf with a brass globe on top dark bookshelves present, no brass globe
  • △ a striped blue-and-yellow rug round grey concentric-ring rug, not striped blue-and-yellow
  • △ a chalkboard easel with a sun drawn on it chalkboard easel present but blank, no sun drawn
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, stands talking with Mr Penguin, a penguin with a black back, white belly, an orange beak, orange webbed feet, and wide curious eyes, the two of them together on the rug beside the chalkboard easel, shown in a wide shot that takes in the whole corner. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page4-night
page4-nightnano-banana
  • ✓ a round window with rain outside
  • △ a red beanbag chair blue beanbag, not red
  • △ a tall oak bookshelf with a brass globe on top bookshelf present, no globe
  • ✗ a striped blue-and-yellow rug no rug shown
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the cozy reading corner. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

nano-banana -- B -- full setting sheet

$0.039 / imagecross-page consistency4night geometry4night palette5same-room pairs3/3unprompted text1

Bookshelf switches sides depending on camera angle (right in p1/p4, left in p2, center-right in p3) -- consistent with rotation, not a true drift. Book spine palette changes: muted browns/greys in p1/p2/p4 vs saturated red/blue/green/yellow in p3. A table lamp appears next to the bookshelf in p4 that is not shown in the day pages. Window size varies (very large in p1, smaller in p2/p3). Chalkboard sun drawing is consistent throughout. Rug stripes consistent.

page1-read
page1-readnano-banana
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • ✓ a tall oak bookshelf with a brass globe on top
  • ✓ a striped blue-and-yellow rug
  • ✓ a chalkboard easel with a sun drawn on it
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page2-search
page2-searchnano-banana
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • ✓ a tall oak bookshelf with a brass globe on top
  • ✓ a striped blue-and-yellow rug
  • ✓ a chalkboard easel with a sun drawn on it
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, kneels at the foot of the tall bookshelf searching the lower shelves for a lost little brass key, shown from a low angle that emphasizes the bookshelf. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page3-talk
page3-talknano-banana
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top books are now brightly multicolored vs muted tones on other pages
  • ✓ a striped blue-and-yellow rug
  • ✓ a chalkboard easel with a sun drawn on it
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, stands talking with Mr Penguin, a penguin with a black back, white belly, an orange beak, orange webbed feet, and wide curious eyes, the two of them together on the rug beside the chalkboard easel, shown in a wide shot that takes in the whole corner. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page4-night
page4-nightnano-banana
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • ✓ a tall oak bookshelf with a brass globe on top
  • ✓ a striped blue-and-yellow rug
  • ✓ a chalkboard easel with a sun drawn on it
promptThe reading corner: a cozy indoor nook with a round window on one wall showing steady rain outside, a plump red beanbag chair on the floor, a tall oak bookshelf standing against the wall with a polished brass globe resting on top, a striped blue-and-yellow rug covering the floor, and a chalkboard easel in the corner with a simple yellow sun drawn on it in chalk. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

nano-banana -- C -- compact 3-anchor lock

$0.039 / imagecross-page consistency3night geometry3night palette5same-room pairs3/3unprompted text0

Brass globe on top of bookshelf is missing on all four pages. Striped blue-and-yellow rug is missing on every page (floor is gray on 1-2, bluish speckled on 3, dark on 4). Chalkboard easel with sun only appears on page 3, and even there the sun drawing is absent (board is blank/green). Window frame style varies: plain wooden ring (pages 1, 2, 4) vs. muntined cross frame (page 3). Bookshelf switches wall side between pages (right on 1, 3, 4; left on 2) which is acceptable as camera change but book colors shift from vivid multicolor (page 1) to muted brown (pages 2-4). Page 4 introduces a side table and lamp not seen elsewhere.

page1-read
page1-readnano-banana
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top tall bookshelf present but no brass globe on top
  • ✗ a striped blue-and-yellow rug plain gray floor, no rug
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page2-search
page2-searchnano-banana
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelf present, no globe
  • ✗ a striped blue-and-yellow rug bare gray floor
  • ✗ a chalkboard easel with a sun drawn on it not visible
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, kneels at the foot of the tall bookshelf searching the lower shelves for a lost little brass key, shown from a low angle that emphasizes the bookshelf. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page3-talk
page3-talknano-banana
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelf present, no globe on top
  • △ a striped blue-and-yellow rug bluish speckled floor, not striped blue-and-yellow
  • △ a chalkboard easel with a sun drawn on it chalkboard easel present but blank, no sun drawing
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is daytime, with soft grey light coming through the rainy window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, stands talking with Mr Penguin, a penguin with a black back, white belly, an orange beak, orange webbed feet, and wide curious eyes, the two of them together on the rug beside the chalkboard easel, shown in a wide shot that takes in the whole corner. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
page4-night
page4-nightnano-banana
  • ✓ a round window with rain outside
  • ✓ a red beanbag chair
  • △ a tall oak bookshelf with a brass globe on top bookshelf visible, no globe
  • ✗ a striped blue-and-yellow rug dark bare floor
  • ✗ a chalkboard easel with a sun drawn on it replaced by side table with lamp
promptThe scene takes place in the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf. The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room. It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window. Maria, a fifth-grade Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, a dark grey jacket worn open over an orange top, and blue jeans, sits curled in the beanbag reading a picture book by the window, shown in a medium shot. bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

Full written results

the complete document behind this page -- method, every table, the reads, and the standing caveats

_Last updated: 2026-08-04_

Experiment 4 -- setting continuity results (prompt-only, no reference images)

Where stable-image-core sits in this experiment

SIC is not a candidate to render pages, and its cells below are not a shortlist comparison. It is in the round because it is what production renders today, so it sets the baseline a migration has to beat, and because it is the model that mints our style exemplars -- it renders the target picker styles, graphic novel and 3D animation especially, better than anything else we have measured. Experiment 4 tests the other half of the job. The goal is for gpt-image-2 and nano-banana to reach SIC's style quality while also holding the room together and getting the plot right, which is where SIC falls apart: its best prompt-only cell reaches 15% anchor recall, it does not honor the graphic-novel style at all here, and it volunteers lettering in every single cell. Read its rows as the size of the problem. Its seed lever is still measured (arm D) because it is a real property of the incumbent, not because it is a reason to keep it.

What this experiment tested

Two-plus story pages are supposed to happen in the same room. With NO reference image allowed on any model, how close can prompt-side technique get to "recognizably the same room" across pages that recompose it, and which technique wins per model. Style is held fixed at graphic-novel (a controlled variable, not an axis), characters ride as text only, and character drift is not scored -- the room is the whole subject.

Slice run: one setting (the reading corner) x 4 pages x 3 models (stable-image-core, gpt-image-2, nano-banana) x arms A/B/C, plus arm D (seed-hold) on stable-image-core only = 40 renders. n is small everywhere; read directions, not decimals.

The setting sheet (the reading corner)

Five enumerable, judge-checkable anchors, each positively phrased with a shape/position, plus a blank-surface clause baked into the sheet (text-suppression folded into the shared material so it never becomes a confounded arm):

  1. a round window with rain outside
  2. a red beanbag chair
  3. a tall oak bookshelf with a brass globe on top
  4. a striped blue-and-yellow rug
  5. a chalkboard easel with a sun drawn on it

Blank-surface clause: "The walls are bare, the book spines are unlabeled, and there is no signage or lettering anywhere in the room."

The story (four pages, Maria + Mr Penguin, text descriptions only)

The four pages map onto the design's four page pressures:

  • Page 1 (solo action): "Maria settled into the beanbag with a picture book open on her knees and read quietly by the window while the rain tapped the glass."
  • Page 2 (different-area action): "She had lost the little brass key to her diary, so she knelt at the foot of the bookshelf and searched along the lower shelves." (Bookshelf area emphasized, low angle.)
  • Page 3 (two-character interaction, wider framing): "Mr Penguin waddled over across the rug, and the two of them stood together by the easel and talked about where the key might be."
  • Page 4 (nighttime version of the solo action): "Long after dark, Maria curled back into the beanbag by the window and read one last page before bed."

The night page keeps the setting anchor text byte-identical to page 1 and swaps only the lighting clause ("It is daytime, with soft grey light coming through the rainy window." becomes "It is night: the room is dark, lit only by a small warm lamp, and the rain still streaks the round window."), so the page isolates lighting-vs-geometry drift instead of confounding it with reworded scene text.

The arms

  • A -- name-only control: "The scene takes place in the cozy reading corner." plus the blank-surface clause.
  • B -- full setting sheet, verbatim every page: the character-sheet discipline applied to place.
  • C -- compact three-anchor lock: "the reading corner with the round rain-streaked window, the red beanbag chair, and the tall oak bookshelf" plus the blank-surface clause.
  • D -- seed-hold (stable-image-core only): arm B's text with the SAME seed on all four pages; arms A/B/C on stable-image-core derive a different seed per page (production behavior).

The fixed style clause on every arm and page: "bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded."

Results

Judge: opus-4-7 over Bedrock (the only judge comparable with the rest of this program; GPT-4o was NOT substituted). Cross-page consistency 1-5 (5 = unmistakably the same room every page), night geometry and night palette scored separately, unprompted-text a per-cell binary, anchor recall = share of sheet anchors present across the four pages.

ModelArmCross-page consistencyNight geometryNight paletteUnprompted textAnchor recall
stable-image-coreA334yes0%
stable-image-coreB221yes15%
stable-image-coreC215yes15%
stable-image-coreD (seed-hold)551yes20%
gpt-image-2A225no15%
gpt-image-2B345no85%
gpt-image-2C345no40%
nano-bananaA225yes5%
nano-bananaB445yes95%
nano-bananaC335no40%

Per-model read

  • nano-banana produces the single best cell in the round: arm B holds all five anchors on every page (95% recall) and comes back as the same room (cpc 4), with the night page surviving both geometry and palette. It drifts badly without the full sheet (arm A recall 5%, cpc 2). It does volunteer incidental lettering on arms A and B (see the text row) -- the MAYAY-setting risk shows up on the model that markets text rendering as a strength.
  • gpt-image-2 is the same shape one notch lower: arm B recall 85%, cpc 3, night geometry 4, and -- distinctively -- it never volunteers unprompted lettering in any cell. It is the cleanest model on incidental text.
  • stable-image-core (the incumbent) is the weakest on the room and the worst on text: it does not honor the graphic-novel style, its best prompt-only cell (arm B) reaches only 15% recall at cpc 2, and it volunteers chalkboard/wall lettering on every cell. Its arm D (seed-hold) is discussed below.

What replicated or contradicted the design's priors

  • The design's B-vs-C prior did NOT replicate. The documented-idiom prior said the compact three-anchor lock (C) should tie or beat the full sheet (B). It did not: the ordering is B > C > A on both reference models. The compact lock holds its three named anchors but drops the other two (recall ~40% vs B's 85-95%). On cross-call setting continuity, detail density is what carries it -- the full sheet is worth keeping, not trimming.
  • Arm A behaves as the design's null predicted: name-only drifts (cpc 2, recall 0-15%), confirming the round is measuring a real effect.
  • Seed-hold (arm D) confirms the design's predicted tradeoff, degenerately. It maxes consistency (cpc 5, geometry 5) only because the same seed makes all four pages near-duplicate compositions -- the room "matches" because the pages are essentially the same picture, and the night page fails to darken (palette 1). It buys sameness by killing recomposition and breaks the lighting page, so it is not a real production win despite the top-line score. Only stable-image-core can even try it (OpenAI/Gemini expose no seed).
  • The MAYAY-setting text family is real and model-specific: volunteered lettering on every stable-image-core cell and on nano-banana arms A/B, never on gpt-image-2 -- despite the blank-surface clause being in every prompt. Positive blank-surface phrasing did not suppress it on the two text-prone models.
  • The night page held up under arm B on the reference models (geometry 4, palette 5 on both gpt-image-2 B and nano-banana B) -- the least-documented, highest-risk cell survived when the full sheet anchored it.

Standing caveats

  • n = 1 setting, 1 story, 4 pages per cell. Directions only. The round is built to swap in three new settings without touching machinery (the experiment 3b replication move) to catch any sheet-specific artifact before these directions are treated as settled.
  • Character drift across pages is expected and unscored (experiment 4 attaches no references).
  • Style is fixed graphic-novel; results read onto the likely ship style, not onto today's production storybook-default suffix.
  • Anchor recall is pooled across the four pages (present count / total anchor checks); a low number can mean either the anchor never appeared or appeared on only some pages.

Run notes (incidents, all resolved)

Two harness bugs surfaced and were fixed mid-run, and the test-account credential expired overnight between the render pass and the final judge pass:

  1. The runner's env parser keyed the credentials file's export KEY=value lines as "export AWS_ACCESS_KEY_ID", leaving the AWS keys undefined so the Bedrock calls silently fell back to the (expired) SSO cache. Fixed by stripping the export prefix.
  2. Both judges declared image media type image/png, but stable-image-core returns JPEG, so Bedrock rejected every stable-image-core judge call. Fixed by sniffing the media type from the magic bytes. The stable-image-core images were already on disk, so those four cells were re-judged from the saved renders with no re-rendering.

No renders were wasted; the credential was refreshed and every cell is opus-judged.

Story Studio scene accuracy -- experiment 5

Findings
  • When a picture does not match its story, it is almost always the people who are wrong. Sending the story paragraph as-is, both models draw the action and the props nearly perfectly -- gpt-image-2 got 86% of actions and 100% of objects right -- while getting the character wrong three-quarters of the time (24% of character details correct on gpt-image-2, 18% on nano-banana). Naming what each character looks like takes that to 100% on both. Scenery was never the problem.
  • Structured beat prompts (arm B) are worth 18-26 recall points over piping the paragraph in: gpt-image-2 went from 74% recall / 3.6 composition to a PERFECT 100% / 5.0 on both tiers -- including every binding trap; nano-banana 72% to 90%. Binding errors go 22%→0% (gpt-image-2) and 27%→9% (nano).
  • Honest caveat on that number: arm A never states what the characters look like, so arm B bundles attribute injection with labeled structure, explicit counts, and an exclusion clause. The gap sits almost entirely in the actor category, so naming your characters' appearance is likely the dominant component. The ablation that splits them is the next run.
  • Attaching a reference image neither helped nor hurt scene accuracy here. On gpt-image-2 the arm with references attached scored identically to the one without (100% / 5.0); on nano-banana it gained 4 points. The division of labour is clean at this complexity: the reference image establishes who the character looks like, and the prompt text establishes what is happening. (On the much harder tier-5 beats this stops holding -- see the E5b tab, where references cost accuracy.)
  • Stable Image Core cannot assemble a scene. It placed 32% of the scripted elements, attached 45% of them to the wrong character or object, scored 2.31 on composition, and got 0 of 4 spatial relationships right under both prompting styles. Better prompting did not move it at all -- composition was 2.31 either way. That is the practical case for the migration: it draws our styles beautifully and cannot draw the story, so it stays the tool that creates character references and stops being the tool that renders pages.
  • Correcting experiment 4 on lettering: experiment 4's "gpt-image-2 never letters" was an artifact of its blank-surface clause being in every arm. With no exclusion clause gpt-image-2 is the WORST offender -- 14/15 arm-A cells (93%), inventing plausible comic-shop set dressing ("WATCHMEN", "MAUS", "SANDMAN" book spines). The good news: one exclusion sentence is a reliable off-switch, 0/16 in arms B and C on both models. It does NOT work on stable-image-core (38% arm A, 50% arm B).
  • One DETERMINISTIC OpenAI safety block on a benign beat ("Brad tipped his green watering can over the leafy potted plant"): 3 rejections across 2 runs, output-stage, category "other". The program's first benign-content block after 100+ renders. It did NOT reproduce in arms B/C, which rendered the same beat clean -- so reprompting is a viable fallback. That cell is left missing; gpt-image-2 arm A is n=15.
Recommended next steps
  • Refined by experiment 5b -- read that tab before acting on this number. Decomposing this prompt showed the accuracy gain comes from one of its four parts, naming each character's appearance, which the live prompt builder already does. A second part, the no-text clause, is essential for a different reason (it stops invented lettering). The remaining two showed nothing on these beats -- but on the harder tier-5 beats the full structured prompt still beats the bare paragraph 91% to 65%, and that tier has not been decomposed yet.
  • Done (E5b): attribute injection carries recall; the exclusion clause carries lettering; labeled structure and explicit counts earn nothing measurable.
  • Attach character references freely; no accuracy cost measured. Do not expect them to fix plot accuracy.
  • Done (E5b): tier-5 compound traps de-saturated the rubric -- gpt-image-2 drops to 91% recall / 8-of-10 joint-correct, so regressions are detectable again.
  • Run tiers 2-3 to complete the degradation curve, and validate the scripted beats against 10 real production paragraphs.
  • Ship a block-handling path with reprompt-as-fallback; log the benign block in the moderation battery. One specimen is a tally, not a policy.
Read the full written results for this experiment ↓method, every table, the reads, and the standing caveats

Does the picture show what the story says? Element-by-element recall with binding checked -- WHO does WHAT with WHICH object WHERE.

  • The material: a story yielding 8 checklisted beats -- 4 simple (tier 1: one actor, one action, one prop) and 4 binding traps (tier 4: crossed attributes, exact counts, spatial precision, instrument binding) -- cast across Aisha, Brad, Maria, Mr Penguin, and one trio.
  • Arm A -- paragraph-as-prompt: the story paragraph sent as-is. The control floor.
  • Arm B -- structured beat prompt: the same beat recomposed in each model's documented idiom.
  • Arm C -- references attached: arm B plus the characters' frozen graphic-novel references (gpt-image-2 and nano-banana only) -- measures the reference-attention tax.
  • Scoring: every checklist element marked present (✓), misbound (⇆ -- there, but wrong actor/object/place), or absent (✗); plus composition 1-5 and an unprompted-text check. Two replicates per cell.
  • stable-image-core is not a candidate here. It is in the round because it is what production renders today and because it mints our style exemplars -- it renders the target styles better than anything else we have measured. This round tests the other half of the job, which is the half it fails. Read its cells as the size of the problem.
beats 8 (tiers 1 + 4)judge claude-opus-4-7 via Bedrockarms paragraph / structured / +refsruns 2026-08-03

stable-image-core

stable-image-core -- A -- paragraph-as-prompt (control)

$0.04 / imagetier-1 recall50%tier-1 misbound37%tier-4 recall21%tier-4 misbound50%composition avg2.3unprompted text6
t1-aisha-read
t1-aisha-read · r1stable-image-core · tier 1
composition2unprompted text!
  • ⇆ aisha -- green dress, gold hoop earrings, long black hair Woman is present but wears a blue vest and white shirt, not a green dress; hair is short/medium brown, no visible gold hoop earrings.
  • ⇆ Aisha reads the open book while sitting on the stool She is reading a book while seated, but on a wooden chair, not a stool.
  • ✓ open book (count 1)
  • ⇆ wooden stool (count 1) She sits on a wooden chair with a backrest, not a stool.
promptAisha sat on a wooden stool and read from a single open book held in her lap. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-aisha-read
t1-aisha-read · r2stable-image-core · tier 1
composition3unprompted text!
  • ⇆ aisha -- attributes: green dress, gold hoop earrings, long black hair Woman has gold hoop earring and long black hair, but wears a teal/blue outfit and headscarf, not a green dress.
  • ⇆ Aisha reads the open book while sitting on the stool She reads a book but is seated on a chair, not the stool; a stool is separate in the scene.
  • ✓ open book (count 1; no special attributes)
  • ✓ wooden stool (count 1; no special attributes)
promptAisha sat on a wooden stool and read from a single open book held in her lap. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-brad-water
t1-brad-water · r1stable-image-core · tier 1
composition2
  • ✗ brad -- attributes: tan jacket, black trousers, short dark wavy hair No human figure in the image.
  • ✗ Brad waters the potted plant No actor performing the watering action.
  • ✓ watering can (count 1; green)
  • ⇆ potted plant (count 1; leafy) Multiple leafy potted plants shown instead of exactly one.
promptBrad tipped his green watering can over the leafy potted plant on the windowsill and watered it slowly. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-brad-water
t1-brad-water · r2stable-image-core · tier 1
composition2
  • ✗ brad -- tan jacket, black trousers, short dark wavy hair No human figure visible in the image.
  • ✗ Brad waters the potted plant No actor performing the watering action.
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptBrad tipped his green watering can over the leafy potted plant on the windowsill and watered it slowly. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-maria-umbrella
t1-maria-umbrella · r1stable-image-core · tier 1
composition3
  • ⇆ maria -- orange thick-rimmed glasses, grey jacket, short black hair Woman has short dark hair but no orange glasses and wears a red jacket instead of grey.
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptMaria smiled and held one red umbrella up over her head as the first drops fell. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-maria-umbrella
t1-maria-umbrella · r2stable-image-core · tier 1
composition3unprompted text!
  • ⇆ maria -- attributes: orange thick-rimmed glasses, grey jacket, short black hair No glasses; jacket is red not grey; hair is dark reddish-brown not black
  • ⇆ Maria holds the umbrella up over her head Umbrella is above her but no visible hand grasping it
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptMaria smiled and held one red umbrella up over her head as the first drops fell. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-penguin-fish
t1-penguin-fish · r1stable-image-core · tier 1
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptMr Penguin waddled up from the water holding one shiny silver fish in his beak. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-penguin-fish
t1-penguin-fish · r2stable-image-core · tier 1
composition2
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ⇆ Mr Penguin holds the fish in his beak The fish (with bird-like features) appears to be held by the penguin's flipper/wing, not in his beak.
  • ⇆ fish (count 1; silver) One creature is shown but it looks more like a bird with a fish tail than a silver fish.
promptMr Penguin waddled up from the water holding one shiny silver fish in his beak. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r1stable-image-core · tier 4
composition2unprompted text!
  • ⇆ brad -- tan jacket, black trousers Man is present but wearing a blue denim shirt, not a tan jacket; trousers not clearly black.
  • ⇆ Brad stands beside the table Brad appears to be seated at/behind the table, not standing beside it.
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ⇆ cat is UNDER the table, not on top Cat is on top of the table, not underneath -- the exact trap.
  • ✗ Brad: worried, brows drawn up Expression looks flat/stern, not worried with brows drawn up.
promptBrad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r2stable-image-core · tier 4
composition2unprompted text!
  • ⇆ brad -- tan jacket, black trousers Man present but wearing blue shirt, not tan jacket; trousers not visible as black
  • ⇆ Brad stands beside the table Brad is seated at the table, not standing beside it
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ⇆ cat is UNDER the table Cat is on top of the table, not under it
  • ✗ Brad worried, brows drawn up Expression looks flat/stern, not clearly worried with raised brows
promptBrad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r1stable-image-core · tier 4
composition2unprompted text!
  • ⇆ maria -- red scarf, orange thick-rimmed glasses, grey jacket Maria has a red/yellow scarf but no glasses and wears a blue jacket, not grey
  • ⇆ mr-penguin -- blue scarf, orange beak, white belly Penguin has orange beak and white belly but no blue scarf
  • ✗ Maria hands the flower to Mr Penguin Maria holds flower at her side, not handing it over
  • ✗ Mr Penguin receives the flower Penguin is not receiving anything
  • ⇆ scarf on Maria (count 1; red) Scarf is red and yellow, not solely red
  • ✗ scarf on Mr Penguin (count 1; blue) Penguin wears no scarf
  • ✓ flower (count 1; yellow)
promptMaria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r2stable-image-core · tier 4
composition2
  • ⇆ maria -- red scarf, orange thick-rimmed glasses, grey jacket Human character absent; a penguin figure wears the red scarf but has blue jacket and no glasses
  • ⇆ mr-penguin -- blue scarf, orange beak, white belly Penguin has orange beak and white belly but wears a red scarf, not blue
  • ⇆ Maria hands the flower to Mr Penguin The taller penguin-like figure holds a yellow item, but Maria is not present as described
  • ✗ Mr Penguin receives the flower Small penguin is not shown receiving anything
  • ⇆ scarf on Maria (count 1; red) Red scarf is on the wrong actor (a penguin), not Maria
  • ⇆ scarf on Mr Penguin (count 1; blue) Mr Penguin wears a red scarf instead of blue
  • ⇆ flower (count 1; yellow) Many yellow flowers in the scene rather than a single one being handed over
promptMaria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r1stable-image-core · tier 4
composition1
  • ✗ aisha -- green dress, gold hoop earrings No person visible in the image.
  • ✗ Aisha arranges the clay pots on the shelf No actor performing arrangement.
  • ⇆ clay pots on the shelf (count 3) Many pots visible on shelves, not exactly three; count is wrong.
  • ⇆ daisy in the middle pot (count 1; white) Multiple white daisies visible instead of one.
  • ⇆ three pots in a row on shelf, daisy in middle Pots are on multiple shelves, not a clear row of three with daisy centered.
promptAisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r2stable-image-core · tier 4
composition2
  • ✗ aisha -- green dress, gold hoop earrings No person visible in the image.
  • ✗ Aisha arranges the clay pots on the shelf No actor performing the action.
  • ⇆ clay pots on the shelf (count 3) Four vases shown instead of three, and they appear ceramic/glazed rather than plain clay.
  • ⇆ daisy in the middle pot (count 1; white) Multiple daisies appear in several pots, not a single daisy in the middle pot.
  • ⇆ three pots in a row, daisy in the middle Four pots present; daisies distributed across pots, not isolated to middle.
promptAisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r1stable-image-core · tier 4
composition2
  • ✗ brad -- tan jacket, black trousers No human male in tan jacket visible; only a woman and two penguins.
  • ⇆ aisha -- green dress, gold hoop earrings A woman is present but wears a pink/magenta dress with blue coat, no green dress or gold hoops.
  • ✓ mr-penguin -- orange beak, white belly
  • ✗ Brad opens the wooden chest Brad is not in the image; chest appears closed.
  • ⇆ Aisha holds the lantern up The lantern is sitting on the chest; the woman touches it but does not hold it aloft, and a lantern also hangs from ceiling.
  • ⇆ Mr Penguin only watches, does nothing The larger penguin is actively reaching/touching the lantern rather than only watching.
  • ✓ wooden chest (count 1)
  • ⇆ lantern (count 1) Multiple lanterns shown (on chest, hanging from ceiling, on side table) instead of exactly one.
promptBrad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r2stable-image-core · tier 4
composition2
  • ✗ brad -- tan jacket, black trousers No human characters visible, only penguins
  • ✗ aisha -- green dress, gold hoop earrings No human characters visible
  • ✓ mr-penguin -- orange beak, white belly
  • ✗ Brad opens the wooden chest Brad not present; chest appears closed
  • ⇆ Aisha holds the lantern up Lantern is sitting on chest, not held by Aisha (who is absent)
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptBrad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

stable-image-core -- B -- structured beat prompt

$0.04 / imagetier-1 recall50%tier-1 misbound33%tier-4 recall27%tier-4 misbound46%composition avg2.3unprompted text8
t1-aisha-read
t1-aisha-read · r1stable-image-core · tier 1
composition3unprompted text!
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ⇆ Aisha reads the open book while sitting on the stool She is holding a closed book while sitting on a chair, not a stool; an open book rests on her lap.
  • ✓ open book (count 1)
  • ⇆ wooden stool (count 1) She sits on a wooden chair with backrest, not a stool.
promptAisha (green dress, gold hoop earrings, long black hair), Aisha reads the open book while sitting on the stool, exactly one open book, exactly one wooden stool, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t1-aisha-read
t1-aisha-read · r2stable-image-core · tier 1
composition3unprompted text!
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ⇆ Aisha reads the open book while sitting on the stool She is seated on a wooden chair with a back, not a stool; also appears to be writing rather than reading.
  • ✓ open book (count 1)
  • ✓ wooden stool (count 1)
promptAisha (green dress, gold hoop earrings, long black hair), Aisha reads the open book while sitting on the stool, exactly one open book, exactly one wooden stool, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t1-brad-water
t1-brad-water · r1stable-image-core · tier 1
composition2unprompted text!
  • ⇆ brad -- tan jacket, black trousers, short dark wavy hair Jacket is green, not tan; trousers and hair match.
  • ✗ Brad waters the potted plant He is standing and holding an object but not watering; no water shown.
  • ✗ watering can (count 1; green) He holds a small black object, not a green watering can.
  • ✓ potted plant (count 1; leafy)
promptBrad (tan jacket, black trousers, short dark wavy hair), Brad waters the potted plant, exactly one green watering can, exactly one leafy potted plant, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t1-brad-water
t1-brad-water · r2stable-image-core · tier 1
composition2
  • ⇆ brad -- tan jacket, black trousers, short dark wavy hair Has black trousers and dark wavy hair but jacket is green, not tan
  • ✗ Brad waters the potted plant Brad stands with hands in pockets, not watering
  • ✗ watering can (count 1; green) No watering can visible
  • ✓ potted plant (count 1; leafy)
promptBrad (tan jacket, black trousers, short dark wavy hair), Brad waters the potted plant, exactly one green watering can, exactly one leafy potted plant, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t1-maria-umbrella
t1-maria-umbrella · r1stable-image-core · tier 1
composition4unprompted text!
  • ⇆ maria -- orange thick-rimmed glasses, grey jacket, short black hair Grey jacket and short black hair match, but glasses are black-rimmed, not orange.
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptMaria (orange thick-rimmed glasses, grey jacket, short black hair), Maria holds the umbrella up over her head, exactly one red umbrella, Maria looks smiling, mouth turned up, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t1-maria-umbrella
t1-maria-umbrella · r2stable-image-core · tier 1
composition3unprompted text!
  • ⇆ maria -- attributes: orange thick-rimmed glasses, grey jacket, short black hair Grey jacket and short black hair present, but glasses are black-rimmed, not orange
  • ✓ Maria holds the umbrella up over her head
  • ⇆ umbrella (count 1; red) One umbrella, but canopy is mostly grey with red trim, not fully red
  • ✓ Maria: smiling, mouth turned up
promptMaria (orange thick-rimmed glasses, grey jacket, short black hair), Maria holds the umbrella up over her head, exactly one red umbrella, Maria looks smiling, mouth turned up, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t1-penguin-fish
t1-penguin-fish · r1stable-image-core · tier 1
composition2
  • ✓ mr-penguin -- black back, white belly, orange beak, orange webbed feet
  • ✗ Mr Penguin holds the fish in his beak penguin's beak is empty; fish float separately in the air
  • ⇆ fish (count 1; silver) two fish shown (one orange, one silver) instead of a single silver fish
promptMr Penguin (black back, white belly, orange beak, orange webbed feet), Mr Penguin holds the fish in his beak, exactly one silver fish, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t1-penguin-fish
t1-penguin-fish · r2stable-image-core · tier 1
composition3
  • ✓ mr-penguin -- black back, white belly, orange beak, orange webbed feet
  • ⇆ Mr Penguin holds the fish in his beak Fish is held by/under the flipper, not in the beak.
  • ✓ fish (count 1; silver)
promptMr Penguin (black back, white belly, orange beak, orange webbed feet), Mr Penguin holds the fish in his beak, exactly one silver fish, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t4-cat-under-table
t4-cat-under-table · r1stable-image-core · tier 4
composition2unprompted text!
  • ⇆ brad -- tan jacket, black trousers Man present but wears dark grey/black jacket, not tan
  • ⇆ Brad stands beside the table He is seated at the table, not standing beside it
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ⇆ the cat is UNDER the table, not on top of it Cat is sitting ON top of the table
  • ✗ Brad: worried, brows drawn up Expression looks neutral/stern, not worried
promptBrad (tan jacket, black trousers), Brad stands beside the table, exactly one wooden table, exactly one orange cat, the cat is UNDER the table, not on top of it, Brad looks worried: brows drawn up, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t4-cat-under-table
t4-cat-under-table · r2stable-image-core · tier 4
composition2unprompted text!
  • ⇆ brad -- attributes: tan jacket, black trousers Jacket is orange, not tan; trousers not visible.
  • ⇆ Brad stands beside the table Brad appears seated behind the desk, not standing beside it.
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ⇆ the cat is UNDER the table, not on top of it The cat is on top of the table, not under it.
  • ✗ Brad: worried: brows drawn up Expression looks neutral/stern, not worried.
promptBrad (tan jacket, black trousers), Brad stands beside the table, exactly one wooden table, exactly one orange cat, the cat is UNDER the table, not on top of it, Brad looks worried: brows drawn up, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t4-crossed-scarves
t4-crossed-scarves · r1stable-image-core · tier 4
composition2
  • ✗ maria -- red scarf, orange thick-rimmed glasses, grey jacket Maria is not depicted; only a penguin is shown
  • ⇆ mr-penguin -- blue scarf, orange beak, white belly Penguin has orange beak and white belly, but wears an orange/red scarf instead of blue
  • ✗ Maria hands the flower to Mr Penguin No Maria present; penguin holds flower himself
  • ⇆ Mr Penguin receives the flower Penguin holds the flower but is not receiving it from anyone
  • ⇆ scarf on Maria (count 1; red) A red/orange scarf exists but is worn by the penguin, not Maria
  • ⇆ scarf on Mr Penguin (count 1; blue) Penguin wears a scarf, but it is red/orange, not blue
  • ✓ flower (count 1; yellow)
promptMaria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t4-crossed-scarves
t4-crossed-scarves · r2stable-image-core · tier 4
composition1
  • ✗ maria -- red scarf, orange thick-rimmed glasses, grey jacket No human Maria visible in image
  • ⇆ mr-penguin -- blue scarf, orange beak, white belly Penguin present with orange beak and white belly, but scarf is orange (not blue) and glasses are red (not on Maria)
  • ✗ Maria hands the flower to Mr Penguin No handing action; Maria absent and no flower being exchanged
  • ✗ Mr Penguin receives the flower Penguin is not receiving a flower; flowers are background decoration
  • ✗ scarf on Maria (count 1; red) Maria not depicted
  • ⇆ scarf on Mr Penguin (count 1; blue) Scarf on penguin is orange, not blue
  • ⇆ flower (count 1; yellow) Multiple flowers shown in background (yellow and orange), not a single yellow flower being held
promptMaria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t4-three-pots
t4-three-pots · r1stable-image-core · tier 4
composition2
  • ✓ aisha -- green dress, gold hoop earrings
  • ⇆ Aisha arranges the clay pots on the shelf She holds a small pot but is not clearly arranging pots on a shelf; pots are already arranged around her
  • ⇆ clay pots on the shelf (count 3) Many more than 3 clay pots visible on shelves
  • ⇆ daisy in the middle pot (count 1; white) White daisies visible but not clearly in a single middle pot of a row of three
  • ✗ three pots in a row on shelf, daisy in middle No clear row of exactly three pots with daisy in middle
promptAisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, exactly three clay pots on the shelf, exactly one white daisy in the middle pot, the three pots are in a row on the shelf, the daisy in the middle pot, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t4-three-pots
t4-three-pots · r2stable-image-core · tier 4
composition2
  • ✓ aisha -- green dress, gold hoop earrings
  • ✗ Aisha arranges the clay pots on the shelf She stands with hands clasped, not arranging pots
  • ⇆ clay pots on the shelf (count 3) Many more than 3 clay pots shown on shelves and counter
  • ⇆ daisy in the middle pot (count 1, white) Daisies present but in multiple side pots, not clearly one middle pot
  • ✗ three pots in a row on shelf, daisy in middle pot No clear row of exactly three pots with daisy centered
promptAisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, exactly three clay pots on the shelf, exactly one white daisy in the middle pot, the three pots are in a row on the shelf, the daisy in the middle pot, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t4-trio-chest
t4-trio-chest · r1stable-image-core · tier 4
composition2unprompted text!
  • ✗ brad -- tan jacket, black trousers No male character present
  • ⇆ aisha -- green dress, gold hoop earrings Woman has green vest and gold hoops but wears orange coat/pants, not a green dress
  • ✓ mr-penguin -- orange beak, white belly
  • ✗ Brad opens the wooden chest No Brad and no chest being opened
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ⇆ wooden chest (count 1) A briefcase/case is shown instead of a wooden chest
  • ✓ lantern (count 1)
promptBrad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, exactly one wooden chest, exactly one lantern, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering
t4-trio-chest
t4-trio-chest · r2stable-image-core · tier 4
composition2
  • ⇆ brad -- tan jacket, black trousers Male figure wears dark suit with orange vest, not tan jacket.
  • ⇆ aisha -- green dress, gold hoop earrings Green coat over white dress; no visible hoop earrings.
  • ✓ mr-penguin -- orange beak, white belly
  • ✗ Brad opens the wooden chest Chest is closed; Brad stands with hands in pockets.
  • ⇆ Aisha holds the lantern up Lantern is mounted on wall, not held by Aisha.
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ⇆ lantern (count 1) Lantern exists but is a wall sconce, not a portable held lantern.
promptBrad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, exactly one wooden chest, exactly one lantern, a simple bright classroom setting, plain background, bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded, no lettering

gpt-image-2

gpt-image-2 -- A -- paragraph-as-prompt (control)

$0.040 / imagetier-1 recall77%tier-1 misbound23%tier-4 recall73%tier-4 misbound21%composition avg3.6unprompted text14
t1-aisha-read
t1-aisha-read · r1gpt-image-2 · tier 1
composition3unprompted text!
  • ⇆ aisha -- attributes: green dress, gold hoop earrings, long black hair Actor present with gold hoop earrings, but wearing black/purple dress and pink hijab instead of green dress and long black hair
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1; no special attributes)
  • ✓ wooden stool (count 1; no special attributes)
promptAisha sat on a wooden stool and read from a single open book held in her lap. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-aisha-read
t1-aisha-read · r2gpt-image-2 · tier 1
composition3unprompted text!
  • ⇆ aisha -- green dress, gold hoop earrings, long black hair Wears black top with purple hijab, not green dress; gold hoops present but hair is covered.
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1)
  • ✓ wooden stool (count 1)
promptAisha sat on a wooden stool and read from a single open book held in her lap. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-brad-water
t1-brad-water · r1gpt-image-2 · tier 1
composition4unprompted text!
  • ⇆ brad -- tan jacket, black trousers, short dark wavy hair Actor present but wearing green shirt and blue jeans with blond hair, not the specified attributes.
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptBrad tipped his green watering can over the leafy potted plant on the windowsill and watered it slowly. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-maria-umbrella
t1-maria-umbrella · r1gpt-image-2 · tier 1
composition3unprompted text!
  • ⇆ maria -- attributes: orange thick-rimmed glasses, grey jacket, short black hair Maria present but wearing red jacket, no glasses; hair is medium-length not short
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptMaria smiled and held one red umbrella up over her head as the first drops fell. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-maria-umbrella
t1-maria-umbrella · r2gpt-image-2 · tier 1
composition3unprompted text!
  • ⇆ maria -- attributes: orange thick-rimmed glasses, grey jacket, short black hair Maria present but wears no glasses, jacket is tan/beige not grey, hair is long not short
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptMaria smiled and held one red umbrella up over her head as the first drops fell. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-penguin-fish
t1-penguin-fish · r1gpt-image-2 · tier 1
composition5unprompted text!
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptMr Penguin waddled up from the water holding one shiny silver fish in his beak. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-penguin-fish
t1-penguin-fish · r2gpt-image-2 · tier 1
composition4
  • ⇆ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet Penguin present with black back and white belly, but beak is black/dark, not orange, and feet are dark rather than orange.
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptMr Penguin waddled up from the water holding one shiny silver fish in his beak. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r1gpt-image-2 · tier 4
composition4unprompted text!
  • ⇆ brad -- attributes: tan jacket, black trousers Wearing blue jacket and tan trousers, colors swapped
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top of it
  • ✓ Brad: worried: brows drawn up
promptBrad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r2gpt-image-2 · tier 4
composition3unprompted text!
  • ⇆ brad -- attributes: tan jacket, black trousers Man present but wears blue hoodie and blue jeans, not tan jacket/black trousers.
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ⇆ the cat is UNDER the table, not on top of it Cat sits beside/in front of the table, not clearly under it.
  • ✓ Brad: worried: brows drawn up
promptBrad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r1gpt-image-2 · tier 4
composition4unprompted text!
  • ⇆ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket Red scarf present but no glasses and jacket is brown, not grey
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✗ Mr Penguin receives the flower Penguin's flippers are on its own chest, not reaching for flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptMaria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r2gpt-image-2 · tier 4
composition4unprompted text!
  • ⇆ maria -- red scarf, orange thick-rimmed glasses, grey jacket Red scarf and jacket present but no orange thick-rimmed glasses; jacket is blue-teal rather than grey
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptMaria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r1gpt-image-2 · tier 4
composition3unprompted text!
  • ⇆ aisha -- attributes: green dress, gold hoop earrings Woman present with gold earrings but wears a blue dress with purple shawl, not green
  • ✗ Aisha arranges the clay pots on the shelf She stands with arms crossed, looking at pots, not arranging them
  • ✓ clay pots on the shelf (count 3; no special attributes)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ the three pots are in a row on the shelf, the daisy in the middle pot
promptAisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r2gpt-image-2 · tier 4
composition3unprompted text!
  • ⇆ aisha -- attributes: green dress, gold hoop earrings Woman wears teal/blue dress and teardrop earrings, not green dress with gold hoops.
  • ✗ Aisha arranges the clay pots on the shelf She stands with arms crossed, not arranging pots.
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row on shelf, daisy in middle pot
promptAisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r1gpt-image-2 · tier 4
composition4unprompted text!
  • ⇆ brad -- tan jacket, black trousers Brad present but wears white shirt with red scarf and tan trousers, not tan jacket/black trousers.
  • ⇆ aisha -- green dress, gold hoop earrings Aisha present with gold hoop earrings but wears white top and purple sash, not green dress.
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptBrad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r2gpt-image-2 · tier 4
composition4unprompted text!
  • ⇆ brad -- tan jacket, black trousers Brad wears brown/tan jacket but trousers are olive/khaki, not black
  • ⇆ aisha -- green dress, gold hoop earrings Gold hoop earrings present, but wearing blue jacket and khaki pants, not a green dress
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptBrad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

gpt-image-2 -- B -- structured beat prompt

$0.040 / imagetier-1 recall100%tier-1 misbound0%tier-4 recall100%tier-4 misbound0%composition avg5.0unprompted text0
t1-aisha-read
t1-aisha-read · r1gpt-image-2 · tier 1
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1)
  • ✓ wooden stool (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair). Action: Aisha reads the open book while sitting on the stool. Objects: exactly one open book, exactly one wooden stool. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-aisha-read
t1-aisha-read · r2gpt-image-2 · tier 1
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1)
  • ✓ wooden stool (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair). Action: Aisha reads the open book while sitting on the stool. Objects: exactly one open book, exactly one wooden stool. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-brad-water
t1-brad-water · r1gpt-image-2 · tier 1
composition5
  • ✓ brad -- attributes: tan jacket, black trousers, short dark wavy hair
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers, short dark wavy hair). Action: Brad waters the potted plant. Objects: exactly one green watering can, exactly one leafy potted plant. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-brad-water
t1-brad-water · r2gpt-image-2 · tier 1
composition5
  • ✓ brad -- attributes: tan jacket, black trousers, short dark wavy hair
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers, short dark wavy hair). Action: Brad waters the potted plant. Objects: exactly one green watering can, exactly one leafy potted plant. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-maria-umbrella
t1-maria-umbrella · r1gpt-image-2 · tier 1
composition5
  • ✓ maria -- attributes: orange thick-rimmed glasses, grey jacket, short black hair
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (orange thick-rimmed glasses, grey jacket, short black hair). Action: Maria holds the umbrella up over her head. Objects: exactly one red umbrella. Expression: Maria looks smiling, mouth turned up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-maria-umbrella
t1-maria-umbrella · r2gpt-image-2 · tier 1
composition5
  • ✓ maria -- attributes: orange thick-rimmed glasses, grey jacket, short black hair
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (orange thick-rimmed glasses, grey jacket, short black hair). Action: Maria holds the umbrella up over her head. Objects: exactly one red umbrella. Expression: Maria looks smiling, mouth turned up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-penguin-fish
t1-penguin-fish · r1gpt-image-2 · tier 1
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptScene: a simple bright classroom setting, plain background. Subjects: Mr Penguin (black back, white belly, orange beak, orange webbed feet). Action: Mr Penguin holds the fish in his beak. Objects: exactly one silver fish. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-penguin-fish
t1-penguin-fish · r2gpt-image-2 · tier 1
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptScene: a simple bright classroom setting, plain background. Subjects: Mr Penguin (black back, white belly, orange beak, orange webbed feet). Action: Mr Penguin holds the fish in his beak. Objects: exactly one silver fish. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-cat-under-table
t4-cat-under-table · r1gpt-image-2 · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ cat is UNDER the table
  • ✓ Brad worried: brows drawn up
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers). Action: Brad stands beside the table. Objects: exactly one wooden table, exactly one orange cat. Placement: the cat is UNDER the table, not on top of it. Expression: Brad looks worried: brows drawn up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-cat-under-table
t4-cat-under-table · r2gpt-image-2 · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ cat is UNDER the table, not on top
  • ✓ Brad worried: brows drawn up
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers). Action: Brad stands beside the table. Objects: exactly one wooden table, exactly one orange cat. Placement: the cat is UNDER the table, not on top of it. Expression: Brad looks worried: brows drawn up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-crossed-scarves
t4-crossed-scarves · r1gpt-image-2 · tier 4
composition5
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Action: Maria hands the flower to Mr Penguin; Mr Penguin receives the flower. Objects: exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-crossed-scarves
t4-crossed-scarves · r2gpt-image-2 · tier 4
composition5
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Action: Maria hands the flower to Mr Penguin; Mr Penguin receives the flower. Objects: exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-three-pots
t4-three-pots · r1gpt-image-2 · tier 4
composition5
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row, daisy in the middle pot
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings). Action: Aisha arranges the clay pots on the shelf. Objects: exactly three clay pots on the shelf, exactly one white daisy in the middle pot. Placement: the three pots are in a row on the shelf, the daisy in the middle pot. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-three-pots
t4-three-pots · r2gpt-image-2 · tier 4
composition5
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row on the shelf, daisy in the middle pot
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings). Action: Aisha arranges the clay pots on the shelf. Objects: exactly three clay pots on the shelf, exactly one white daisy in the middle pot. Placement: the three pots are in a row on the shelf, the daisy in the middle pot. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-trio-chest
t4-trio-chest · r1gpt-image-2 · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Action: Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing. Objects: exactly one wooden chest, exactly one lantern. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-trio-chest
t4-trio-chest · r2gpt-image-2 · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Action: Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing. Objects: exactly one wooden chest, exactly one lantern. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.

gpt-image-2 -- C -- structured + references attached

$0.040 / imagetier-1 recall100%tier-1 misbound0%tier-4 recall100%tier-4 misbound0%composition avg5.0unprompted text0
t1-aisha-read
t1-aisha-read · r1gpt-image-2 · tier 1 +ref
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1)
  • ✓ wooden stool (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair). Action: Aisha reads the open book while sitting on the stool. Objects: exactly one open book, exactly one wooden stool. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-aisha-read
t1-aisha-read · r2gpt-image-2 · tier 1 +ref
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1)
  • ✓ wooden stool (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair). Action: Aisha reads the open book while sitting on the stool. Objects: exactly one open book, exactly one wooden stool. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-brad-water
t1-brad-water · r1gpt-image-2 · tier 1 +ref
composition5
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers, short dark wavy hair). Action: Brad waters the potted plant. Objects: exactly one green watering can, exactly one leafy potted plant. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-brad-water
t1-brad-water · r2gpt-image-2 · tier 1 +ref
composition5
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers, short dark wavy hair). Action: Brad waters the potted plant. Objects: exactly one green watering can, exactly one leafy potted plant. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-maria-umbrella
t1-maria-umbrella · r1gpt-image-2 · tier 1 +ref
composition5
  • ✓ maria -- orange thick-rimmed glasses, grey jacket, short black hair
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (orange thick-rimmed glasses, grey jacket, short black hair). Action: Maria holds the umbrella up over her head. Objects: exactly one red umbrella. Expression: Maria looks smiling, mouth turned up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-maria-umbrella
t1-maria-umbrella · r2gpt-image-2 · tier 1 +ref
composition5
  • ✓ maria -- orange thick-rimmed glasses, grey jacket, short black hair
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (orange thick-rimmed glasses, grey jacket, short black hair). Action: Maria holds the umbrella up over her head. Objects: exactly one red umbrella. Expression: Maria looks smiling, mouth turned up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-penguin-fish
t1-penguin-fish · r1gpt-image-2 · tier 1 +ref
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptScene: a simple bright classroom setting, plain background. Subjects: Mr Penguin (black back, white belly, orange beak, orange webbed feet). Action: Mr Penguin holds the fish in his beak. Objects: exactly one silver fish. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t1-penguin-fish
t1-penguin-fish · r2gpt-image-2 · tier 1 +ref
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptScene: a simple bright classroom setting, plain background. Subjects: Mr Penguin (black back, white belly, orange beak, orange webbed feet). Action: Mr Penguin holds the fish in his beak. Objects: exactly one silver fish. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-cat-under-table
t4-cat-under-table · r1gpt-image-2 · tier 4 +ref
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ cat is UNDER the table, not on top
  • ✓ Brad worried: brows drawn up
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers). Action: Brad stands beside the table. Objects: exactly one wooden table, exactly one orange cat. Placement: the cat is UNDER the table, not on top of it. Expression: Brad looks worried: brows drawn up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-cat-under-table
t4-cat-under-table · r2gpt-image-2 · tier 4 +ref
composition5
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top of it
  • ✓ Brad: worried: brows drawn up
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers). Action: Brad stands beside the table. Objects: exactly one wooden table, exactly one orange cat. Placement: the cat is UNDER the table, not on top of it. Expression: Brad looks worried: brows drawn up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-crossed-scarves
t4-crossed-scarves · r1gpt-image-2 · tier 4 +ref
composition5
  • ✓ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Action: Maria hands the flower to Mr Penguin; Mr Penguin receives the flower. Objects: exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-crossed-scarves
t4-crossed-scarves · r2gpt-image-2 · tier 4 +ref
composition5
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Action: Maria hands the flower to Mr Penguin; Mr Penguin receives the flower. Objects: exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-three-pots
t4-three-pots · r1gpt-image-2 · tier 4 +ref
composition5
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (white)
  • ✓ three pots in a row, daisy in middle pot
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings). Action: Aisha arranges the clay pots on the shelf. Objects: exactly three clay pots on the shelf, exactly one white daisy in the middle pot. Placement: the three pots are in a row on the shelf, the daisy in the middle pot. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-three-pots
t4-three-pots · r2gpt-image-2 · tier 4 +ref
composition5
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row on shelf, daisy in middle pot
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings). Action: Aisha arranges the clay pots on the shelf. Objects: exactly three clay pots on the shelf, exactly one white daisy in the middle pot. Placement: the three pots are in a row on the shelf, the daisy in the middle pot. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-trio-chest
t4-trio-chest · r1gpt-image-2 · tier 4 +ref
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (1)
  • ✓ lantern (1)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Action: Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing. Objects: exactly one wooden chest, exactly one lantern. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-trio-chest
t4-trio-chest · r2gpt-image-2 · tier 4 +ref
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Action: Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing. Objects: exactly one wooden chest, exactly one lantern. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.

nano-banana

nano-banana -- A -- paragraph-as-prompt (control)

$0.039 / imagetier-1 recall77%tier-1 misbound23%tier-4 recall69%tier-4 misbound29%composition avg3.6unprompted text7
t1-aisha-read
t1-aisha-read · r1nano-banana · tier 1
composition3unprompted text!
  • ⇆ aisha -- attributes: green dress, gold hoop earrings, long black hair Woman present but wearing blue/white shirt and jeans, no green dress, no gold hoop earrings, hair is curly not long straight black
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1; no special attributes)
  • ✓ wooden stool (count 1; no special attributes)
promptAisha sat on a wooden stool and read from a single open book held in her lap. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-aisha-read
t1-aisha-read · r2nano-banana · tier 1
composition3
  • ⇆ aisha -- attributes: green dress, gold hoop earrings, long black hair Woman is present but wears a blue shirt and yellow/gold skirt, not a green dress; no visible hoop earrings; hair is curly shoulder-length rather than long.
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1; no special attributes)
  • ✓ wooden stool (count 1; no special attributes)
promptAisha sat on a wooden stool and read from a single open book held in her lap. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-brad-water
t1-brad-water · r1nano-banana · tier 1
composition4unprompted text!
  • ⇆ brad -- attributes: tan jacket, black trousers, short dark wavy hair Man has short dark hair but wears a blue sweater, not a tan jacket; trousers not visible.
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptBrad tipped his green watering can over the leafy potted plant on the windowsill and watered it slowly. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-brad-water
t1-brad-water · r2nano-banana · tier 1
composition4unprompted text!
  • ⇆ brad -- attributes: tan jacket, black trousers, short dark wavy hair Man present but wears blue t-shirt and tan trousers, no tan jacket or black trousers; hair is light brown, not dark.
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptBrad tipped his green watering can over the leafy potted plant on the windowsill and watered it slowly. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-maria-umbrella
t1-maria-umbrella · r1nano-banana · tier 1
composition3unprompted text!
  • ⇆ maria -- orange thick-rimmed glasses, grey jacket, short black hair Woman present but wearing yellow hoodie (not grey jacket) and no orange glasses; hair is medium length.
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptMaria smiled and held one red umbrella up over her head as the first drops fell. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-maria-umbrella
t1-maria-umbrella · r2nano-banana · tier 1
composition3
  • ⇆ maria -- orange thick-rimmed glasses, grey jacket, short black hair Woman present but wearing yellow raincoat, long wavy hair, no glasses
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptMaria smiled and held one red umbrella up over her head as the first drops fell. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-penguin-fish
t1-penguin-fish · r1nano-banana · tier 1
composition4
  • ⇆ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet Penguin has black back and white belly but beak and feet are black/dark, not orange.
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptMr Penguin waddled up from the water holding one shiny silver fish in his beak. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t1-penguin-fish
t1-penguin-fish · r2nano-banana · tier 1
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptMr Penguin waddled up from the water holding one shiny silver fish in his beak. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r1nano-banana · tier 4
composition4
  • ⇆ brad -- attributes: tan jacket, black trousers Man is present but wears a blue shirt and grey trousers, not tan jacket/black trousers.
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top of it
  • ✓ Brad: worried: brows drawn up
promptBrad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r2nano-banana · tier 4
composition4
  • ⇆ brad -- attributes: tan jacket, black trousers Man is present but wearing a blue shirt, not a tan jacket; trousers are gray, not black.
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top of it
  • ✓ Brad: worried: brows drawn up
promptBrad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r1nano-banana · tier 4
composition4
  • ⇆ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket Has red scarf but no orange glasses and jacket is blue, not grey
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptMaria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r2nano-banana · tier 4
composition4
  • ⇆ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket Has red scarf but no orange glasses and jacket is blue, not grey
  • ⇆ mr-penguin -- attributes: blue scarf, orange beak, white belly Blue scarf and white belly present, but beak is black not orange
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptMaria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r1nano-banana · tier 4
composition2unprompted text!
  • ⇆ aisha -- green dress, gold hoop earrings Woman wears red dress with green scarf, not green dress; no visible gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ⇆ clay pots on the shelf (count 3) Only 2 clay pots visible on the shelf, not 3
  • ⇆ daisy in the middle pot (count 1; white) White daisy is present but in one of two pots; middle position ambiguous without three pots
  • ⇆ three pots in a row, daisy in the middle pot Only two pots shown, so there is no true middle pot
promptAisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r2nano-banana · tier 4
composition3
  • ⇆ aisha -- attributes: green dress, gold hoop earrings Woman is present in green top but no visible gold hoop earrings; hair is white/silver.
  • ✗ Aisha arranges the clay pots on the shelf She is not shown touching or arranging the pots; hands not visible.
  • ✓ clay pots on the shelf (count 3; no special attributes)
  • ⇆ daisy in the middle pot (count 1; white) A white daisy appears in the middle pot but there is also an extra dark flower alongside it.
  • ✓ the three pots are in a row on the shelf, the daisy in the middle pot
promptAisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r1nano-banana · tier 4
composition4unprompted text!
  • ⇆ brad -- attributes: tan jacket, black trousers Man wears brown tunic and tan/olive trousers, not a tan jacket with black trousers.
  • ⇆ aisha -- attributes: green dress, gold hoop earrings Woman wears blue/purple dress, not green; earring visible but dress color wrong.
  • ✓ mr-penguin -- attributes: orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptBrad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r2nano-banana · tier 4
composition4unprompted text!
  • ⇆ brad -- tan jacket, black trousers Male figure present but wearing red shirt, not tan jacket
  • ⇆ aisha -- green dress, gold hoop earrings Wears green jacket but no visible gold hoop earrings, and it's not a dress
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptBrad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

nano-banana -- B -- structured beat prompt

$0.039 / imagetier-1 recall97%tier-1 misbound3%tier-4 recall87%tier-4 misbound12%composition avg4.6unprompted text0
t1-aisha-read
t1-aisha-read · r1nano-banana · tier 1
composition5
  • ✓ aisha -- attributes: green dress, gold hoop earrings, long black hair
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1; no special attributes)
  • ✓ wooden stool (count 1; no special attributes)
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair), Aisha reads the open book while sitting on the stool, with exactly one open book, exactly one wooden stool, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-aisha-read
t1-aisha-read · r2nano-banana · tier 1
composition5
  • ✓ aisha -- attributes: green dress, gold hoop earrings, long black hair
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1; no special attributes)
  • ✓ wooden stool (count 1; no special attributes)
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair), Aisha reads the open book while sitting on the stool, with exactly one open book, exactly one wooden stool, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-brad-water
t1-brad-water · r1nano-banana · tier 1
composition4
  • ✓ brad -- attributes: tan jacket, black trousers, short dark wavy hair
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ⇆ potted plant (count 1; leafy) Two leafy potted plants visible instead of one.
promptA graphic-novel illustration of Brad (tan jacket, black trousers, short dark wavy hair), Brad waters the potted plant, with exactly one green watering can, exactly one leafy potted plant, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-brad-water
t1-brad-water · r2nano-banana · tier 1
composition5
  • ✓ brad -- attributes: tan jacket, black trousers, short dark wavy hair
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptA graphic-novel illustration of Brad (tan jacket, black trousers, short dark wavy hair), Brad waters the potted plant, with exactly one green watering can, exactly one leafy potted plant, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-maria-umbrella
t1-maria-umbrella · r1nano-banana · tier 1
composition5
  • ✓ maria -- attributes: orange thick-rimmed glasses, grey jacket, short black hair
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptA graphic-novel illustration of Maria (orange thick-rimmed glasses, grey jacket, short black hair), Maria holds the umbrella up over her head, with exactly one red umbrella, Maria looks smiling, mouth turned up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-maria-umbrella
t1-maria-umbrella · r2nano-banana · tier 1
composition5
  • ✓ maria -- orange thick-rimmed glasses, grey jacket, short black hair
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptA graphic-novel illustration of Maria (orange thick-rimmed glasses, grey jacket, short black hair), Maria holds the umbrella up over her head, with exactly one red umbrella, Maria looks smiling, mouth turned up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-penguin-fish
t1-penguin-fish · r1nano-banana · tier 1
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptA graphic-novel illustration of Mr Penguin (black back, white belly, orange beak, orange webbed feet), Mr Penguin holds the fish in his beak, with exactly one silver fish, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-penguin-fish
t1-penguin-fish · r2nano-banana · tier 1
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptA graphic-novel illustration of Mr Penguin (black back, white belly, orange beak, orange webbed feet), Mr Penguin holds the fish in his beak, with exactly one silver fish, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-cat-under-table
t4-cat-under-table · r1nano-banana · tier 4
composition5
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top of it
  • ✓ Brad: worried: brows drawn up
promptA graphic-novel illustration of Brad (tan jacket, black trousers), Brad stands beside the table, with exactly one wooden table, exactly one orange cat, the cat is UNDER the table, not on top of it, Brad looks worried: brows drawn up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-cat-under-table
t4-cat-under-table · r2nano-banana · tier 4
composition3
  • ✓ brad -- tan jacket, black trousers
  • ⇆ Brad stands beside the table Brad stands behind the table with hands on it, not clearly beside it.
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ⇆ cat is UNDER the table, not on top Cat is on the chair, not under the table.
  • ✓ Brad worried: brows drawn up
promptA graphic-novel illustration of Brad (tan jacket, black trousers), Brad stands beside the table, with exactly one wooden table, exactly one orange cat, the cat is UNDER the table, not on top of it, Brad looks worried: brows drawn up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-crossed-scarves
t4-crossed-scarves · r1nano-banana · tier 4
composition4
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ⇆ Maria hands the flower to Mr Penguin Maria holds the flower but Mr Penguin is not reaching to receive; direction of transfer is ambiguous/reversed.
  • ✗ Mr Penguin receives the flower Penguin's flippers are not extended to receive; Maria is the one holding it.
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptA graphic-novel illustration of Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, with exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-crossed-scarves
t4-crossed-scarves · r2nano-banana · tier 4
composition5
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptA graphic-novel illustration of Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, with exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-three-pots
t4-three-pots · r1nano-banana · tier 4
composition5
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row on shelf, daisy in middle pot
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, with exactly three clay pots on the shelf, exactly one white daisy in the middle pot, the three pots are in a row on the shelf, the daisy in the middle pot, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-three-pots
t4-three-pots · r2nano-banana · tier 4
composition3
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ⇆ clay pots on the shelf (count 3) Only 2 clay pots are visible on the shelf, not 3
  • ⇆ daisy in the middle pot (count 1; white) A white daisy is present but it's in one of only two pots, not a middle pot of three
  • ⇆ three pots in a row on shelf, daisy in middle Only two pots shown; middle-of-three arrangement fails
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, with exactly three clay pots on the shelf, exactly one white daisy in the middle pot, the three pots are in a row on the shelf, the daisy in the middle pot, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-trio-chest
t4-trio-chest · r1nano-banana · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptA graphic-novel illustration of Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, with exactly one wooden chest, exactly one lantern, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-trio-chest
t4-trio-chest · r2nano-banana · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptA graphic-novel illustration of Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, with exactly one wooden chest, exactly one lantern, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.

nano-banana -- C -- structured + references attached

$0.039 / imagetier-1 recall100%tier-1 misbound0%tier-4 recall90%tier-4 misbound10%composition avg4.7unprompted text0
t1-aisha-read
t1-aisha-read · r1nano-banana · tier 1 +ref
composition5
  • ✓ aisha -- attributes: green dress, gold hoop earrings, long black hair
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1; no special attributes)
  • ✓ wooden stool (count 1; no special attributes)
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair), Aisha reads the open book while sitting on the stool, with exactly one open book, exactly one wooden stool, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-aisha-read
t1-aisha-read · r2nano-banana · tier 1 +ref
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha reads the open book while sitting on the stool
  • ✓ open book (count 1)
  • ✓ wooden stool (count 1)
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair), Aisha reads the open book while sitting on the stool, with exactly one open book, exactly one wooden stool, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-brad-water
t1-brad-water · r1nano-banana · tier 1 +ref
composition5
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptA graphic-novel illustration of Brad (tan jacket, black trousers, short dark wavy hair), Brad waters the potted plant, with exactly one green watering can, exactly one leafy potted plant, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-brad-water
t1-brad-water · r2nano-banana · tier 1 +ref
composition5
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad waters the potted plant
  • ✓ watering can (count 1; green)
  • ✓ potted plant (count 1; leafy)
promptA graphic-novel illustration of Brad (tan jacket, black trousers, short dark wavy hair), Brad waters the potted plant, with exactly one green watering can, exactly one leafy potted plant, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-maria-umbrella
t1-maria-umbrella · r1nano-banana · tier 1 +ref
composition5
  • ✓ maria -- attributes: orange thick-rimmed glasses, grey jacket, short black hair
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptA graphic-novel illustration of Maria (orange thick-rimmed glasses, grey jacket, short black hair), Maria holds the umbrella up over her head, with exactly one red umbrella, Maria looks smiling, mouth turned up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-maria-umbrella
t1-maria-umbrella · r2nano-banana · tier 1 +ref
composition5
  • ✓ maria -- orange thick-rimmed glasses, grey jacket, short black hair
  • ✓ Maria holds the umbrella up over her head
  • ✓ umbrella (count 1; red)
  • ✓ Maria: smiling, mouth turned up
promptA graphic-novel illustration of Maria (orange thick-rimmed glasses, grey jacket, short black hair), Maria holds the umbrella up over her head, with exactly one red umbrella, Maria looks smiling, mouth turned up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-penguin-fish
t1-penguin-fish · r1nano-banana · tier 1 +ref
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptA graphic-novel illustration of Mr Penguin (black back, white belly, orange beak, orange webbed feet), Mr Penguin holds the fish in his beak, with exactly one silver fish, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t1-penguin-fish
t1-penguin-fish · r2nano-banana · tier 1 +ref
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak, orange webbed feet
  • ✓ Mr Penguin holds the fish in his beak
  • ✓ fish (count 1; silver)
promptA graphic-novel illustration of Mr Penguin (black back, white belly, orange beak, orange webbed feet), Mr Penguin holds the fish in his beak, with exactly one silver fish, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-cat-under-table
t4-cat-under-table · r1nano-banana · tier 4 +ref
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top
  • ✓ Brad: worried, brows drawn up
promptA graphic-novel illustration of Brad (tan jacket, black trousers), Brad stands beside the table, with exactly one wooden table, exactly one orange cat, the cat is UNDER the table, not on top of it, Brad looks worried: brows drawn up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-cat-under-table
t4-cat-under-table · r2nano-banana · tier 4 +ref
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ cat is UNDER the table
  • ✓ Brad worried: brows drawn up
promptA graphic-novel illustration of Brad (tan jacket, black trousers), Brad stands beside the table, with exactly one wooden table, exactly one orange cat, the cat is UNDER the table, not on top of it, Brad looks worried: brows drawn up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-crossed-scarves
t4-crossed-scarves · r1nano-banana · tier 4 +ref
composition4
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ⇆ Mr Penguin receives the flower Penguin's flipper is extended but not yet grasping the flower; reception ambiguous.
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptA graphic-novel illustration of Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, with exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-crossed-scarves
t4-crossed-scarves · r2nano-banana · tier 4 +ref
composition5
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptA graphic-novel illustration of Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, with exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-three-pots
t4-three-pots · r1nano-banana · tier 4 +ref
composition3
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ⇆ clay pots on the shelf (count 3) Four pots visible on the shelf, not three.
  • ✓ daisy in the middle pot (count 1; white)
  • ⇆ three pots in a row on shelf, daisy in middle pot Pots are in a row but there are four, so 'middle' is ambiguous; count breaks the arrangement.
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, with exactly three clay pots on the shelf, exactly one white daisy in the middle pot, the three pots are in a row on the shelf, the daisy in the middle pot, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-three-pots
t4-three-pots · r2nano-banana · tier 4 +ref
composition3
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ⇆ clay pots on the shelf (count 3) Four pots visible on the shelf, not three
  • ✓ daisy in the middle pot (count 1; white)
  • ⇆ three pots in a row on shelf, daisy in middle pot Pots are in a row but count is 4; daisy is not strictly in the middle of the row
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, with exactly three clay pots on the shelf, exactly one white daisy in the middle pot, the three pots are in a row on the shelf, the daisy in the middle pot, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-trio-chest
t4-trio-chest · r1nano-banana · tier 4 +ref
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptA graphic-novel illustration of Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, with exactly one wooden chest, exactly one lantern, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-trio-chest
t4-trio-chest · r2nano-banana · tier 4 +ref
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptA graphic-novel illustration of Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, with exactly one wooden chest, exactly one lantern, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.

Full written results

the complete document behind this page -- method, every table, the reads, and the standing caveats

_Last updated: 2026-08-04_

Experiment 5 -- scene accuracy results (does the picture show what the story says)

What this experiment tested

Given a story beat -- specific actors doing a specific thing with specific objects in a specific place -- how much of it arrives in the image, correctly bound (WHO does WHAT with WHICH object WHERE), and how accuracy degrades as beats get more complex. Each beat carries a machine-checkable checklist; the judge marks every element present / absent / present-but-misbound, plus a 1-5 composition score. Style is held fixed at graphic-novel across all arms.

Slice run: tier 1 (simple: one actor, one action, one prop) and tier 4 (the binding traps) = 8 beats x 3 models (stable-image-core, gpt-image-2, nano-banana) x arms A/B x 2 replicates = 96, plus arm C (references attached) on gpt-image-2 + nano-banana across the same 8 beats x 2 replicates = 32, for 128 renders total. n is small everywhere; read directions, not decimals.

Where stable-image-core sits in this experiment

Stable Image Core is not a candidate here and its numbers are not a shortlist comparison. It is in the round for two reasons: it is what production renders today, so it sets the baseline anyone migrating away from has to beat, and it is the model that mints our style exemplars, because it renders the target picker styles -- graphic novel and 3D animation especially -- better than anything else we have measured. This round tests the other half of the job. The goal for gpt-image-2 and nano-banana is to reach SIC's style quality while also getting the plot right and putting every named character on the page, which is precisely where SIC falls apart. Read its rows below as the size of the problem, not as a contender being scored.

The story (eight beats, one paragraph each)

Cast spread across the beats: Aisha solo, Brad solo, Maria + Mr Penguin, and a Brad + Aisha + Mr Penguin trio trap. Aisha has no solo reference (her only frozen reference is the brad-aisha pair image); her text description is used in every arm and, in arm C, her pair image is attached with the record flagged pair-only.

Tier 1 (simple):

  • t1-brad-water: "Brad tipped his green watering can over the leafy potted plant on the windowsill and watered it slowly."
  • t1-aisha-read: "Aisha sat on a wooden stool and read from a single open book held in her lap."
  • t1-maria-umbrella: "Maria smiled and held one red umbrella up over her head as the first drops fell." (scripted emotion: smiling)
  • t1-penguin-fish: "Mr Penguin waddled up from the water holding one shiny silver fish in his beak."

Tier 4 (binding traps):

  • t4-crossed-scarves (crossed attributes + instrument binding): "Maria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf." (red->Maria, blue->penguin; Maria gives, penguin receives)
  • t4-three-pots (exact count): "Aisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy."
  • t4-cat-under-table (spatial precision): "Brad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it." (scripted emotion: worried, brows drawn up)
  • t4-trio-chest (instrument binding): "Brad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched."

The arms

  • A -- paragraph-as-prompt control: the story paragraph above, sent as-is (plus the fixed style clause).
  • B -- structured beat prompt in each model's documented idiom: labeled scene->subject->action->object->placement->style segments with an exclusion clause for the gpt family; single-image narrative text in the documented element order (Subject, Composition, Action, Location, Style) with semantic positives and NO first/then/finally step decomposition for Gemini; caption-style for stable-image-core.
  • C -- arm B + the graphic-novel frozen reference(s) for the beat's characters attached (gpt-image-2 and nano-banana only; stable-image-core has no reference path). Arm-C records also carry the judge's binary artifact/quality flag, logged separately from element recall so the reference tax's quality component is not misattributed to a prompt-following deficit.

Results

Judge: opus-4-7 over Bedrock on every cell (the only judge comparable with the rest of this program; GPT-4o was deliberately not substituted). 127 of the 128 designed cells are judged. The one missing cell is t1-brad-water / gpt-image-2 / arm A / replicate 2, lost to a repeatable OpenAI safety rejection -- see the moderation note below.

Element recall counts an element as landed only when it is present AND correctly bound. Misbound means the element is in the picture but attached to the wrong actor, object, or place, which is scored separately from absent because the two failures need different fixes.

Pooled across both tiers:

ModelArmRecallMisboundAbsentCompositionCells with lettering
gpt-image-2A -- paragraph74%22%4%3.6014/15 (93%)
gpt-image-2B -- structured100%0%0%5.000/16
gpt-image-2C -- structured + refs100%0%0%5.000/16
nano-bananaA -- paragraph72%27%1%3.627/16 (44%)
nano-bananaB -- structured90%9%1%4.620/16
nano-bananaC -- structured + refs94%6%0%4.690/16
stable-image-coreA -- paragraph32%45%23%2.316/16 (38%)
stable-image-coreB -- structured35%41%23%2.318/16 (50%)

Split by tier, to read the degradation curve:

ModelArmTier 1 recallTier 4 recallTier 1 compositionTier 4 composition
gpt-image-2A77%73%3.573.62
gpt-image-2B100%100%5.005.00
gpt-image-2C100%100%5.005.00
nano-bananaA77%69%3.623.62
nano-bananaB97%87%4.884.38
nano-bananaC100%90%5.004.38
stable-image-coreA50%21%2.751.88
stable-image-coreB50%27%2.751.88

By element category, which is where the interesting result lives (element counts in parentheses):

ModelArmActorActionObjectSpatialEmotion
gpt-image-2A24% (21)86% (21)100% (28)75% (4)100% (4)
gpt-image-2B100% (22)100% (22)100% (30)100% (4)100% (4)
gpt-image-2C100% (22)100% (22)100% (30)100% (4)100% (4)
nano-bananaA18% (22)95% (22)90% (30)75% (4)100% (4)
nano-bananaB100% (22)86% (22)90% (30)50% (4)100% (4)
nano-bananaC100% (22)95% (22)93% (30)50% (4)100% (4)
stable-image-coreA18% (22)14% (22)57% (30)0% (4)50% (4)
stable-image-coreB36% (22)23% (22)47% (30)0% (4)50% (4)

The reads

1. Characters are the failure, not scenery. The category table localizes the whole effect. In arm A both reference models render the action and the props almost perfectly -- gpt-image-2 gets 86% of actions and 100% of objects -- and then get the character wrong 76-82% of the time. In arm B actor recall goes to 100% on both. Every other category was already close to fine. So "the illustration does not match the story" is, in this data, overwhelmingly "the people in it are not the right people."

2. Arm B eliminates binding errors on gpt-image-2 and nearly does on nano. Misbinding, the failure the tier-4 traps were built to provoke, goes 22% to 0% on gpt-image-2 and 27% to 9% on nano-banana. gpt-image-2's arm B and C are clean sweeps: 100% recall and 5.00 composition on all 32 cells, tier 1 and tier 4 alike, including all four binding traps.

3. The reference tax measured zero, and so did the reference benefit. Arm C matches arm B exactly on gpt-image-2 (100%/5.00 both) and adds four points on nano-banana (90% to 94%). Attaching character references costs nothing in scene accuracy at this complexity. It also does not fix scene accuracy: the prompt does that. Pair this with experiment 3b, where references were the dominant variable for identity -- references carry who the character looks like, prompt text carries what happens.

4. Tier 4 did not break gpt-image-2, which contradicts the design's prior. The GenEval 2 prior predicted a joint-correctness cliff once counting and spatial atoms co-occur. gpt-image-2 under arm B shows no cliff and no slope: 100% on both tiers. Nano-banana does degrade in the predicted direction (97% to 87%, composition 4.88 to 4.38), so the prior holds for it. Report gpt-image-2's result as a failed prediction rather than a confirmed capability claim, because the rubric saturated (see the caveat below) and n is 2 replicates.

5. Nano-banana's residual weakness is counting and spatial precision, not identity. Once arm B injects attributes it holds actors perfectly, but spatial recall drops from 75% in arm A to 50% in arms B and C, and its worst beat by a wide margin is the count trap t4-three-pots (30% in arm A, 70% in B, 60% in C). Its action recall also dips slightly under arm B (95% to 86%), which is the attention-dilution pattern experiment 3b saw with long nano prompts showing up again on a different axis.

6. Stable Image Core cannot compose a scene, and structuring does not rescue it. 32% recall with 45% of elements misbound, and arm B moves it 3 points while its composition average does not move at all (2.31 both arms). On the tier-4 traps it lands 21-27% of elements. It scored 0 of 4 spatial elements in both arms. This is the clearest statement yet of the split that defines the migration: SIC renders the styles we want and cannot render the story, which is exactly why it stays in the pipeline as the exemplar mint and comes out of it as the page renderer.

What arm A versus arm B actually measures, stated honestly

The 26-point jump is real but it is a bundle, and the largest single component is almost certainly the least sophisticated one. Arm A is the story paragraph as written: "Brad tipped his green watering can over the leafy potted plant on the windowsill and watered it slowly." That sentence never says what Brad looks like. Arm B injects the attribute list. So arm B differs from arm A in at least four ways at once:

  1. character attributes are stated at all (arm A never states them)
  2. the prompt is segmented into labeled fields
  3. counts and spatial relations are restated explicitly ("exactly one red scarf on Maria")
  4. an exclusion clause is appended ("keep every surface free of lettering")

Given that the entire recall gap falls in the actor category, component 1 is doing most of the work, and the round cannot separate it from components 2 through 4. That does not weaken the production recommendation, since production controls all four, but it does mean the honest headline is "naming your characters' appearance in the page prompt is worth about 25 points of scene accuracy" rather than "structured prompts are worth 25 points." The ablation that separates them is the obvious next run, at roughly a dollar.

The judge's reasoning, verbatim, on the sharpest contrast

Same beat (t4-crossed-scarves, the crossed-attribute trap), same judge, three models and arms.

Worst, stable-image-core arm B, composition 2 -- with the full attribute list in the prompt:

"Only one actor is shown and the scarf color-to-actor binding is inverted, missing the key exchange action."

Its element notes: [absent] maria -- "Maria is not depicted; only a penguin is shown", [misbound] scarf on Maria (count 1; red) -- "A red/orange scarf exists but is worn by the penguin, not Maria".

gpt-image-2 arm A, composition 4 -- close, and wrong in the characteristic way:

"Most elements correctly bound with correct colors; Maria missing glasses and jacket color is off, and penguin isn't actively receiving."

gpt-image-2 arm B, composition 5:

"All elements are clearly present and correctly bound in a coherent exchange."

Every element note in that cell reads "All attributes match."

Unprompted lettering: a correction to the experiment 4 finding

Experiment 4 concluded that gpt-image-2 "never volunteers unprompted lettering in any cell." That is an artifact of experiment 4 carrying its blank-surface clause in every arm. With no exclusion clause, gpt-image-2 is the worst text offender in this experiment: 14 of 15 arm-A cells (93%), rising to 8 of 8 on tier 4. It does not scribble noise, it invents plausible comic-shop set dressing. One arm-A cell earned this note:

"Posters read 'MINOR THREAT', 'DESCENDENTS EVERYTHING SUCKS', mug says 'READ MORE COMICS', book spines say 'WATCHMEN', 'MAUS', 'SANDMAN', 'BATMAN', and a sign reads 'Keep Growing Good Things'."

The corrected finding is better news than the original: the exclusion clause is a reliable off-switch on both reference models, 0 of 16 cells in arms B and C on each. The graphic-novel style clause is what invites lettering, and one sentence turns it off. Stable-image-core is the exception and the clause does not work on it at all -- it letters in 38% of arm-A cells and 50% of arm-B cells, going the wrong direction.

Moderation: the program's first benign-content block

t1-brad-water ("Brad tipped his green watering can over the leafy potted plant on the windowsill and watered it slowly") was rejected by OpenAI's safety system on gpt-image-2, arm A, three times: once in the original pass and twice more in a dedicated retry run. All three are HTTP 400 moderation_blocked, moderation_stage: output, categories: ["other"]. It is deterministic, not transient, and it did NOT reproduce on the same beat in arms B or C, both of which rendered and scored clean.

This is the first benign-content block in this program after 100+ renders including every "Disney/Pixar" style arm. Two things follow. Production needs a block-handling path, because a page that cannot render is worse than a page rendered imperfectly, and arms B/C rendering the same beat cleanly suggests reprompting is a viable fallback rather than a dead end. And the block is an output-stage decision on a prompt with nothing objectionable in it, so it cannot be predicted by prompt screening.

The affected cell is left missing rather than substituted. gpt-image-2 arm A is therefore n=15 cells, not 16.

Mint-QC finding (before the slice)

Two representative tier-4 beats were rendered on gpt-image-2 to confirm the props depict: the count trap (t4-three-pots) rendered exactly three clay pots in a row with a single white daisy in the middle pot and Aisha's attributes correct; the spatial trap (t4-cat-under-table) rendered Brad worried (brows up) beside the table with the orange cat clearly UNDER it. Both in graphic-novel style, no incidental lettering. The beats are well-formed and depictable (n=1 each -- a QC confirmation, not a scored result).

Priors to score against (from the design, as tests not baselines)

  • The gpt predecessor's GenEval profile (attribute binding 0.61 < position 0.75 < counting 0.85) predicts the tier-4 difficulty ordering: binding traps hardest, then spatial, then counting.
  • GenEval 2 on gemini-2.5-flash-image measured attributes 91.4%, spatial 70.2%, counting 69.2%, and -- critically -- 84.4% atom-level vs 31.0% prompt-level (all atoms jointly correct). The prediction is a joint-correctness cliff, not a slope, once counting/spatial atoms co-occur.
  • gpt-image-2's press-claimed count-verifying "reasoning" appears in no primary doc, so its counting results are reported as a TEST of that claim, not measured against it. (The n=1 QC render of t4-three-pots got the count right, which is a single anecdote, not evidence for the claim.)

Standing caveats

  • n = 2 replicates per (beat, model, arm), 8 beats over 2 tiers. Directions only; the middle tiers (2 interaction, 3 composed) are a later slice, so the degradation curve has a hole in the middle.
  • The rubric saturated on gpt-image-2. All 32 of its arm-B and arm-C cells scored exactly 100% recall and exactly 5.00 composition. The metric can no longer separate a good render from an excellent one on this model, cannot detect a regression, and cannot show how much headroom is left. Any follow-up that needs to measure gpt-image-2 needs harder beats or a finer scale.
    • Addressed 2026-08-04. A fifth tier of compound traps now exists and a five-cell probe broke the ceiling: 88% pooled recall, 3 of 5 cells jointly correct, composition spanning 2 to 5 on the same model and arm. The probe also found that the judge itself miscounts above roughly five items, which puts the count-and-ordinal beat on hold pending a deterministic checker. Experiment 5's numbers are unaffected: its only count is three, inside the judge's reliable range, and all 133 prompts recorded here still rebuild byte-identically. See 2026-08-04_tier5-harder-beats-rubric-saturation.md.
  • Arm A versus arm B is a four-component bundle, not a clean test of prompt structure -- see the honesty section above. The recall gap is localized almost entirely in the actor category, so stating character attributes is the likely dominant component.
  • Arm C on Aisha beats attaches the brad-aisha PAIR reference (she has no solo reference); records flag it pair-only, so her arm-C identity signal is weaker for that reason alone.
  • Style fixed graphic-novel; results read onto the likely ship style.
  • gpt-image-2 arm A is n=15 cells rather than 16, one cell lost to the moderation block.
  • Element-category subtotals for spatial and emotion rest on 4 elements per cell group. Treat those columns as anecdote.

Run incidents (all resolved)

  1. Credential expiry mid-run. The test-account credential expired between the render pass and the judge pass, blocking the stable-image-core renders and the opus judge (both Bedrock). The gpt-image-2 and nano-banana arms had already rendered with judging deferred, so no render hit a dead judge and nothing was wasted. After the refresh, the deferred cells were judged from the saved renders and the stable-image-core arms were rendered and judged inline. Every cell in the tables above is opus-judged.
  2. Two harness bugs, both fixed during the experiment 4 pass and inherited clean here: the credentials file parser was keying export KEY=value lines with the export prefix attached, leaving AWS keys undefined and silently falling back to an expired SSO cache; and both judges hard-declared image/png while stable-image-core returns JPEG, so Bedrock rejected every stable-image-core judge call. The parser now strips the prefix and the media type is sniffed from magic bytes.
  3. One deterministic moderation block, documented in its own section above.

Every record carries its exact prompt, so both arms are reproducible from the run dirs alone.

What this sets up

  1. The ablation that splits the bundle. Four cells on the tier-4 beats: paragraph plus attributes only, attributes plus labeled structure, plus explicit counts and spatial restatement, plus the exclusion clause. This tells production which parts of arm B to actually build, and no other high-value run on the board costs less.
  2. Harder beats for gpt-image-2, since the current rubric cannot see it any more.
  3. Tiers 2 and 3 to fill the middle of the degradation curve.
  4. Validate against real production paragraphs. All eight beats here were written for the round. Ten real story paragraphs would confirm the effect survives contact with the actual generator's output.
  5. A nano-specific short-form arm on the counting and spatial traps, given the action and spatial dips under the longer structured prompt.
  6. A block-handling policy for the moderation family, with reprompting as the first fallback since arms B and C rendered the same blocked beat cleanly.

Story Studio scene accuracy -- experiment 5b: ablation and compound traps

Findings
  • What was tested. A "structured beat prompt" rewrites a story sentence into labeled parts before it goes to the image model: it names each character's appearance, labels actor / action / object / setting, states counts and positions explicitly, and appends a clause forbidding text in the picture. That measured 20-27 points better than sending the story sentence as-is -- but it is four changes at once, so this ladder adds them back one at a time to find which ones earn their place.
  • Naming each character's appearance carries the accuracy gain. Rung 1 does only that and already reaches 100% -- every scripted element present and correctly attached, on every image, on gpt-image-2. Adding labels (rung 2) and explicit counts (rung 3) moves nothing beyond noise. That is on these beats; the harder tier below has not been taken apart the same way yet.
  • The no-text clause is the lettering component, and it is essential. It adds nothing to accuracy, which is why it looks worthless in the recall column -- but gpt-image-2 invents lettering in 6-8 of every 8 images on every rung without it (speech bubbles, fake book titles, classroom posters) and 0 of 8 with it. Different job, equally necessary.
  • Read the scope limit before acting on this. The ladder ran on tier-4 beats -- one scene, a handful of elements, roughly what a story page asks for today. On the tier-5 compound traps below, the full structured prompt still beats the bare paragraph 91% to 65%. Tier 5 was never decomposed, so we do not yet know whether that 26-point gap is also just character-naming, or whether the labeling starts earning its keep once a scene gets dense.
  • Tier 5 also de-saturated the rubric. gpt-image-2 had been scoring exactly 100% on every structured cell, which looks like success and actually means the test could no longer detect a regression. The harder beats bring it to 91%, and they surfaced the program's first measured downside to attaching reference images: on the densest scenes arm C scores below arm B, 84% against 91%.
Recommended next steps
  • Do not build the full structured-beat assembler on this evidence alone. Of its four components, the one carrying the accuracy gain -- injecting each character's description -- is already in the live image-prompt builder on every page. Build the rest only if the decomposed tier-5 run says they earn it.
  • Add the no-text clause to the production prompt template. One sentence, not there today, and it is the difference between most images carrying invented lettering and none of them doing so. Independent of every other decision on this page.
  • Run this same ladder on the tier-5 beats. That is the missing measurement, and it decides whether labeled structure is inert or merely unnecessary at today's page complexity. Roughly 64 renders.
  • Attach references for character likeness, but expect no scene-accuracy help and a measurable cost on the densest pages.
  • Land the two production prompt defects before comparing anything against live production prompts, or the defects get measured as the arm difference.
Read the full written results for this experiment ↓method, every table, the reads, and the standing caveats

Two follow-ups that experiment 5 forced: which part of structured prompting actually works, and a harder beat set after the rubric saturated.

  • The ablation ladder: arm B changed four things at once. Four cumulative rungs -- attributes, then labeled structure, then explicit counts and spatial relations, then the exclusion clause -- isolate where the gain first appears. Rung 4 is byte-identical to experiment 5 arm B, pinned by a committed fixture, so these numbers are directly comparable rather than merely similar.
  • Tier 5: gpt-image-2 scored a perfect 100% / 5.00 on every structured tier-1/4 cell, so the rubric could no longer see a regression. These five beats require several atoms to be right simultaneously -- high counts crossed with ordinal position, required absence, three-way crossed attributes, comparative size with ordering, handedness with partial framing.
  • Joint-correct is the metric to watch here: the share of renders where EVERY checklist element is right at once. It stays sensitive long after per-element recall pins at 100%.
ladder 4 rungs x 4 traps x 2 models x 2 repstier 5 5 compound beatsjudge claude-opus-4-7 via Bedrockruns 2026-08-05

The ablation ladder

each rung adds ONE component to the one before, on the tier-4 binding traps

gpt-image-2

gpt-image-2 · rung 1 -- attribute injection only

$0.040 / imageelement recall100%misbound0%composition5.0joint-correct8/8unprompted text6
t4-cat-under-table
t4-cat-under-table · r1gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top of it
  • ✓ Brad: worried: brows drawn up
promptBrad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it. Brad (tan jacket, black trousers). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r2gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top of it
  • ✓ Brad: worried: brows drawn up
promptBrad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it. Brad (tan jacket, black trousers). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r1gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptMaria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf. Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r2gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptMaria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf. Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r1gpt-image-2 · tier 4
composition5
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3; no special attributes)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ the three pots are in a row on the shelf, the daisy in the middle pot
promptAisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy. Aisha (green dress, gold hoop earrings). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r2gpt-image-2 · tier 4
composition5
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3; no special attributes)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ the three pots are in a row on the shelf, the daisy in the middle pot
promptAisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy. Aisha (green dress, gold hoop earrings). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r1gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptBrad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched. Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r2gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptBrad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched. Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

gpt-image-2 · rung 2 -- + labeled structure

$0.040 / imageelement recall90%misbound6%composition4.38joint-correct5/8unprompted text8
t4-cat-under-table
t4-cat-under-table · r1gpt-image-2 · tier 4
composition3unprompted text!
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ⇆ the cat is UNDER the table, not on top of it Cat is on top of the table, not under it.
  • ✗ Brad: worried, brows drawn up He appears calm/slightly smiling, not worried.
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers). Action: Brad stands beside the table. Objects: wooden table, orange cat. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r2gpt-image-2 · tier 4
composition3unprompted text!
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ⇆ the cat is UNDER the table, not on top of it Cat is on top of the table, inverted spatial relation.
  • ✗ Brad: worried, brows drawn up He looks calm/slightly smiling, not worried.
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers). Action: Brad stands beside the table. Objects: wooden table, orange cat. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r1gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Action: Maria hands the flower to Mr Penguin; Mr Penguin receives the flower. Objects: red scarf on Maria, blue scarf on Mr Penguin, yellow flower. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r2gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Action: Maria hands the flower to Mr Penguin; Mr Penguin receives the flower. Objects: red scarf on Maria, blue scarf on Mr Penguin, yellow flower. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r1gpt-image-2 · tier 4
composition4unprompted text!
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ⇆ clay pots on the shelf (count 3; no special attributes) Top shelf has 3 pots, but additional pots appear on lower shelf; count on shelf overall exceeds 3.
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ the three pots are in a row on the shelf, the daisy in the middle pot
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings). Action: Aisha arranges the clay pots on the shelf. Objects: clay pots on the shelf, white daisy in the middle pot. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r2gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3; no special attributes)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ the three pots are in a row on the shelf, the daisy in the middle pot
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings). Action: Aisha arranges the clay pots on the shelf. Objects: clay pots on the shelf, white daisy in the middle pot. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r1gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Action: Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing. Objects: wooden chest, lantern. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r2gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ mr-penguin -- attributes: orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1; no special attributes)
  • ✓ lantern (count 1; no special attributes)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Action: Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing. Objects: wooden chest, lantern. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

gpt-image-2 · rung 3 -- + explicit counts & spatial

$0.040 / imageelement recall100%misbound0%composition5.0joint-correct8/8unprompted text6
t4-cat-under-table
t4-cat-under-table · r1gpt-image-2 · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ cat is UNDER the table
  • ✓ Brad: worried, brows drawn up
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers). Action: Brad stands beside the table. Objects: exactly one wooden table, exactly one orange cat. Placement: the cat is UNDER the table, not on top of it. Expression: Brad looks worried: brows drawn up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r2gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ cat is UNDER the table
  • ✓ Brad worried: brows drawn up
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers). Action: Brad stands beside the table. Objects: exactly one wooden table, exactly one orange cat. Placement: the cat is UNDER the table, not on top of it. Expression: Brad looks worried: brows drawn up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r1gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Action: Maria hands the flower to Mr Penguin; Mr Penguin receives the flower. Objects: exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r2gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Action: Maria hands the flower to Mr Penguin; Mr Penguin receives the flower. Objects: exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r1gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3; no special attributes)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ the three pots are in a row on the shelf, the daisy in the middle pot
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings). Action: Aisha arranges the clay pots on the shelf. Objects: exactly three clay pots on the shelf, exactly one white daisy in the middle pot. Placement: the three pots are in a row on the shelf, the daisy in the middle pot. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r2gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row on the shelf, daisy in the middle pot
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings). Action: Aisha arranges the clay pots on the shelf. Objects: exactly three clay pots on the shelf, exactly one white daisy in the middle pot. Placement: the three pots are in a row on the shelf, the daisy in the middle pot. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r1gpt-image-2 · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Action: Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing. Objects: exactly one wooden chest, exactly one lantern. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r2gpt-image-2 · tier 4
composition5unprompted text!
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Action: Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing. Objects: exactly one wooden chest, exactly one lantern. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

gpt-image-2 · rung 4 -- + exclusion clause (= experiment 5 arm B)

$0.040 / imageelement recall98%misbound2%composition4.88joint-correct7/8unprompted text0
t4-cat-under-table
t4-cat-under-table · r1gpt-image-2 · tier 4
composition5
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top of it
  • ✓ Brad: worried: brows drawn up
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers). Action: Brad stands beside the table. Objects: exactly one wooden table, exactly one orange cat. Placement: the cat is UNDER the table, not on top of it. Expression: Brad looks worried: brows drawn up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-cat-under-table
t4-cat-under-table · r2gpt-image-2 · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ cat UNDER the table
  • ✓ Brad worried: brows drawn up
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers). Action: Brad stands beside the table. Objects: exactly one wooden table, exactly one orange cat. Placement: the cat is UNDER the table, not on top of it. Expression: Brad looks worried: brows drawn up. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-crossed-scarves
t4-crossed-scarves · r1gpt-image-2 · tier 4
composition4
  • ✓ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ⇆ Mr Penguin receives the flower Penguin faces flower but flippers do not clearly reach out to receive it.
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Action: Maria hands the flower to Mr Penguin; Mr Penguin receives the flower. Objects: exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-crossed-scarves
t4-crossed-scarves · r2gpt-image-2 · tier 4
composition5
  • ✓ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Action: Maria hands the flower to Mr Penguin; Mr Penguin receives the flower. Objects: exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-three-pots
t4-three-pots · r1gpt-image-2 · tier 4
composition5
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row on shelf, daisy in middle pot
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings). Action: Aisha arranges the clay pots on the shelf. Objects: exactly three clay pots on the shelf, exactly one white daisy in the middle pot. Placement: the three pots are in a row on the shelf, the daisy in the middle pot. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-three-pots
t4-three-pots · r2gpt-image-2 · tier 4
composition5
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row on shelf, daisy in middle pot
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings). Action: Aisha arranges the clay pots on the shelf. Objects: exactly three clay pots on the shelf, exactly one white daisy in the middle pot. Placement: the three pots are in a row on the shelf, the daisy in the middle pot. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-trio-chest
t4-trio-chest · r1gpt-image-2 · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Action: Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing. Objects: exactly one wooden chest, exactly one lantern. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t4-trio-chest
t4-trio-chest · r2gpt-image-2 · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Action: Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing. Objects: exactly one wooden chest, exactly one lantern. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.

nano-banana

nano-banana · rung 1 -- attribute injection only

$0.039 / imageelement recall87%misbound10%composition4.12joint-correct3/8unprompted text1
t4-cat-under-table
t4-cat-under-table · r1nano-banana · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ cat is UNDER the table, not on top
  • ✓ Brad worried, brows drawn up
promptBrad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it. Brad (tan jacket, black trousers). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r2nano-banana · tier 4
composition5
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top of it
  • ✓ Brad: worried: brows drawn up
promptBrad stood beside the wooden table looking worried, his brows drawn up, while his orange cat sat underneath the table, not on top of it. Brad (tan jacket, black trousers). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r1nano-banana · tier 4
composition4
  • ✓ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ⇆ Mr Penguin receives the flower Penguin faces the flower but has no visible hands/flippers reaching to accept it.
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptMaria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf. Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r2nano-banana · tier 4
composition4
  • ✓ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ⇆ Mr Penguin receives the flower Penguin faces the flower but is not shown reaching or grasping it; Maria still holds it.
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptMaria, wearing a red scarf, handed a yellow flower to Mr Penguin, who was wearing a blue scarf. Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r1nano-banana · tier 4
composition4
  • ✓ aisha -- green dress, gold hoop earrings
  • ⇆ Aisha arranges the clay pots on the shelf She touches only the middle pot; not clearly arranging all three, and surface reads as a table rather than a shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row, daisy in the middle
promptAisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy. Aisha (green dress, gold hoop earrings). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r2nano-banana · tier 4
composition3
  • ✓ aisha -- green dress, gold hoop earrings
  • ✗ Aisha arranges the clay pots on the shelf She stands beside the shelf, not touching or arranging pots.
  • ⇆ clay pots on the shelf (count 3) Three terracotta-style pots visible, but a fourth dark pot appears behind them, making count ambiguous.
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row on shelf, daisy in middle pot
promptAisha lined up exactly three clay pots in a row on the wooden shelf, and the middle pot held a single white daisy. Aisha (green dress, gold hoop earrings). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r1nano-banana · tier 4
composition3
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ⇆ Brad opens the wooden chest The penguin's flipper is on the chest lid opening it; Brad is kneeling beside but not clearly opening the lid.
  • ✓ Aisha holds the lantern up
  • ✗ Mr Penguin only watches, does nothing Penguin is actively touching/opening the chest lid, not merely watching.
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptBrad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched. Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r2nano-banana · tier 4
composition5unprompted text!
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptBrad opened the wooden chest while Aisha held the lantern up to light it, and Mr Penguin only watched. Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly). Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

nano-banana · rung 2 -- + labeled structure

$0.039 / imageelement recall81%misbound15%composition3.75joint-correct2/8unprompted text3
t4-cat-under-table
t4-cat-under-table · r1nano-banana · tier 4
composition3
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ⇆ the cat is UNDER the table, not on top of it Cat is sitting ON the table, not under it.
  • ✗ Brad: worried: brows drawn up Expression is neutral/stern, not worried.
promptA graphic-novel illustration of Brad (tan jacket, black trousers), Brad stands beside the table, with wooden table, orange cat, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r2nano-banana · tier 4
composition3
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ⇆ the cat is UNDER the table, not on top of it Cat is on top of the table, not underneath.
  • ✗ Brad: worried: brows drawn up Expression looks neutral/stern, not worried.
promptA graphic-novel illustration of Brad (tan jacket, black trousers), Brad stands beside the table, with wooden table, orange cat, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r1nano-banana · tier 4
composition4
  • ✓ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ⇆ Mr Penguin receives the flower Penguin faces the flower but does not visibly reach out or take it; no flipper contact.
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptA graphic-novel illustration of Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, with red scarf on Maria, blue scarf on Mr Penguin, yellow flower, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r2nano-banana · tier 4
composition4unprompted text!
  • ✓ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ⇆ Mr Penguin receives the flower Penguin faces Maria but does not appear to reach for/take the flower; Maria still holds it.
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptA graphic-novel illustration of Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, with red scarf on Maria, blue scarf on Mr Penguin, yellow flower, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r1nano-banana · tier 4
composition3
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ⇆ clay pots on the shelf (count 3; no special attributes) There are 3 pots on the top shelf but additional 3 pots on the lower shelf, and top shelf count matches 3.
  • ✓ daisy in the middle pot (count 1; white)
  • ⇆ the three pots are in a row on the shelf, the daisy in the middle pot Pots are clustered rather than in a clean row; daisy is in the front pot, not clearly the middle of a row.
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, with clay pots on the shelf, white daisy in the middle pot, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r2nano-banana · tier 4
composition3
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ⇆ clay pots on the shelf (count 3; no special attributes) More than 3 clay pots visible on the shelving unit (at least 5-6 pots across shelves).
  • ✓ daisy in the middle pot (count 1; white)
  • ⇆ the three pots are in a row on the shelf, the daisy in the middle pot Pots are spread across multiple shelves rather than three in a single row; daisy pot is on top shelf but not clearly flanked by exactly two others in a row.
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, with clay pots on the shelf, white daisy in the middle pot, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r1nano-banana · tier 4
composition5unprompted text!
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptA graphic-novel illustration of Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, with wooden chest, lantern, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r2nano-banana · tier 4
composition5unprompted text!
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptA graphic-novel illustration of Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, with wooden chest, lantern, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

nano-banana · rung 3 -- + explicit counts & spatial

$0.039 / imageelement recall92%misbound6%composition4.75joint-correct6/8unprompted text0
t4-cat-under-table
t4-cat-under-table · r1nano-banana · tier 4
composition5
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ✓ the cat is UNDER the table, not on top of it
  • ✓ Brad: worried: brows drawn up
promptA graphic-novel illustration of Brad (tan jacket, black trousers), Brad stands beside the table, with exactly one wooden table, exactly one orange cat, the cat is UNDER the table, not on top of it, Brad looks worried: brows drawn up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-cat-under-table
t4-cat-under-table · r2nano-banana · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1)
  • ✓ cat (count 1; orange)
  • ✓ cat is UNDER the table, not on top
  • ✓ Brad worried: brows drawn up
promptA graphic-novel illustration of Brad (tan jacket, black trousers), Brad stands beside the table, with exactly one wooden table, exactly one orange cat, the cat is UNDER the table, not on top of it, Brad looks worried: brows drawn up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r1nano-banana · tier 4
composition4
  • ✓ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ⇆ Maria hands the flower to Mr Penguin The flower appears to be held by Mr Penguin, with Maria's hands reaching toward it -- looks more like Maria receiving than giving.
  • ⇆ Mr Penguin receives the flower Mr Penguin appears to be holding/offering the flower rather than receiving it.
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptA graphic-novel illustration of Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, with exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-crossed-scarves
t4-crossed-scarves · r2nano-banana · tier 4
composition4
  • ⇆ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket Glasses are present but appear white/light-rimmed, not orange.
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✗ Mr Penguin receives the flower Penguin's wings are down; not reaching for or receiving the flower.
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptA graphic-novel illustration of Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, with exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r1nano-banana · tier 4
composition5
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3; no special attributes)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ the three pots are in a row on the shelf, the daisy in the middle pot
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, with exactly three clay pots on the shelf, exactly one white daisy in the middle pot, the three pots are in a row on the shelf, the daisy in the middle pot, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-three-pots
t4-three-pots · r2nano-banana · tier 4
composition5
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ three pots in a row on shelf, daisy in middle pot
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, with exactly three clay pots on the shelf, exactly one white daisy in the middle pot, the three pots are in a row on the shelf, the daisy in the middle pot, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r1nano-banana · tier 4
composition5
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ mr-penguin -- attributes: orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1; no special attributes)
  • ✓ lantern (count 1; no special attributes)
promptA graphic-novel illustration of Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, with exactly one wooden chest, exactly one lantern, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t4-trio-chest
t4-trio-chest · r2nano-banana · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptA graphic-novel illustration of Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, with exactly one wooden chest, exactly one lantern, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

nano-banana · rung 4 -- + exclusion clause (= experiment 5 arm B)

$0.039 / imageelement recall88%misbound12%composition4.5joint-correct4/8unprompted text0
t4-cat-under-table
t4-cat-under-table · r1nano-banana · tier 4
composition4
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ⇆ the cat is UNDER the table, not on top of it Cat appears to be lying on a lower shelf/support of the table rather than on the floor beneath it, but it is below the top surface.
  • ✓ Brad: worried: brows drawn up
promptA graphic-novel illustration of Brad (tan jacket, black trousers), Brad stands beside the table, with exactly one wooden table, exactly one orange cat, the cat is UNDER the table, not on top of it, Brad looks worried: brows drawn up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-cat-under-table
t4-cat-under-table · r2nano-banana · tier 4
composition4
  • ✓ brad -- attributes: tan jacket, black trousers
  • ✓ Brad stands beside the table
  • ✓ wooden table (count 1; no special attributes)
  • ✓ cat (count 1; orange)
  • ⇆ the cat is UNDER the table, not on top of it Cat is under the table but sitting ON the chair seat, not on the floor beneath
  • ✓ Brad: worried: brows drawn up
promptA graphic-novel illustration of Brad (tan jacket, black trousers), Brad stands beside the table, with exactly one wooden table, exactly one orange cat, the cat is UNDER the table, not on top of it, Brad looks worried: brows drawn up, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-crossed-scarves
t4-crossed-scarves · r1nano-banana · tier 4
composition5
  • ✓ maria -- attributes: red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- attributes: blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ⇆ Mr Penguin receives the flower Penguin faces flower but does not extend flipper to receive it.
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptA graphic-novel illustration of Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, with exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-crossed-scarves
t4-crossed-scarves · r2nano-banana · tier 4
composition5
  • ✓ maria -- red scarf, orange thick-rimmed glasses, grey jacket
  • ✓ mr-penguin -- blue scarf, orange beak, white belly
  • ✓ Maria hands the flower to Mr Penguin
  • ✓ Mr Penguin receives the flower
  • ✓ scarf on Maria (count 1; red)
  • ✓ scarf on Mr Penguin (count 1; blue)
  • ✓ flower (count 1; yellow)
promptA graphic-novel illustration of Maria (red scarf, orange thick-rimmed glasses, grey jacket) and Mr Penguin (blue scarf, orange beak, white belly), Maria hands the flower to Mr Penguin; Mr Penguin receives the flower, with exactly one red scarf on Maria, exactly one blue scarf on Mr Penguin, exactly one yellow flower, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-three-pots
t4-three-pots · r1nano-banana · tier 4
composition5
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ✓ clay pots on the shelf (count 3)
  • ✓ daisy in the middle pot (count 1; white)
  • ✓ the three pots are in a row on the shelf, the daisy in the middle pot
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, with exactly three clay pots on the shelf, exactly one white daisy in the middle pot, the three pots are in a row on the shelf, the daisy in the middle pot, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-three-pots
t4-three-pots · r2nano-banana · tier 4
composition3
  • ✓ aisha -- attributes: green dress, gold hoop earrings
  • ✓ Aisha arranges the clay pots on the shelf
  • ⇆ clay pots on the shelf (count 3; no special attributes) Only 2 clay pots clearly visible on the shelf; count is wrong
  • ⇆ daisy in the middle pot (count 1; white) White daisy is present but in the left/front pot, not the middle of three
  • ⇆ the three pots are in a row on the shelf, the daisy in the middle pot Only two pots in a row; daisy not in a middle position
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings), Aisha arranges the clay pots on the shelf, with exactly three clay pots on the shelf, exactly one white daisy in the middle pot, the three pots are in a row on the shelf, the daisy in the middle pot, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-trio-chest
t4-trio-chest · r1nano-banana · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptA graphic-novel illustration of Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, with exactly one wooden chest, exactly one lantern, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t4-trio-chest
t4-trio-chest · r2nano-banana · tier 4
composition5
  • ✓ brad -- tan jacket, black trousers
  • ✓ aisha -- green dress, gold hoop earrings
  • ✓ mr-penguin -- orange beak, white belly
  • ✓ Brad opens the wooden chest
  • ✓ Aisha holds the lantern up
  • ✓ Mr Penguin only watches, does nothing
  • ✓ wooden chest (count 1)
  • ✓ lantern (count 1)
promptA graphic-novel illustration of Brad (tan jacket, black trousers) and Aisha (green dress, gold hoop earrings) and Mr Penguin (orange beak, white belly), Brad opens the wooden chest; Aisha holds the lantern up; Mr Penguin only watches, does nothing, with exactly one wooden chest, exactly one lantern, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.

Tier 5 -- compound traps

beats requiring several atoms to be right AT ONCE, built after the tier-1/4 rubric saturated

gpt-image-2

gpt-image-2 · arm A -- paragraph as-is (control)

$0.040 / imageelement recall65%misbound34%composition2.7joint-correct0/10unprompted text7
t5-crossed-hats-trio
t5-crossed-hats-trio · r1gpt-image-2 · tier 5
composition3unprompted text!
  • ⇆ maria -- green hat, orange thick-rimmed glasses, grey jacket Green hat present but no orange glasses and jacket is green, not grey.
  • ⇆ aisha -- purple hat, gold hoop earrings, green dress Purple hijab present but no visible gold hoop earrings and wearing purple top, not green dress.
  • ⇆ brad -- orange hat, tan jacket, black trousers Orange cap present but wearing orange t-shirt (no tan jacket) and tan/khaki trousers, not black.
  • ✓ Maria, Aisha, and Brad stand in a row facing the viewer, each holding one book
  • ✓ hat on Maria (count 1; green)
  • ✓ hat on Aisha (count 1; purple)
  • ✓ hat on Brad (count 1; orange)
  • ✓ book held by Maria (count 1; purple)
  • ✓ book held by Aisha (count 1; orange)
  • ✓ book held by Brad (count 1; green)
  • ✓ NO character's book matches their own hat: every one of the three carries a book in a different colour from the hat on their head
promptMaria wore the green hat, Aisha wore the purple hat, and Brad wore the orange hat -- and each of them carried a book in somebody else's colour: Maria's book was purple, Aisha's was orange, and Brad's was green. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t5-crossed-hats-trio
t5-crossed-hats-trio · r2gpt-image-2 · tier 5
composition3unprompted text!
  • ⇆ maria -- green hat, orange thick-rimmed glasses, grey jacket green hat present but no orange glasses and jacket is green, not grey
  • ⇆ aisha -- purple hat, gold hoop earrings, green dress purple hat present but no visible gold hoops and wears purple jacket, not green dress
  • ⇆ brad -- orange hat, tan jacket, black trousers orange hat and dark trousers present but jacket is orange, not tan
  • ⇆ three stand in a row facing viewer, each holding one book three in a row each with one book, but they face each other/turned rather than clearly facing viewer
  • ✓ hat on Maria (green)
  • ✓ hat on Aisha (purple)
  • ✓ hat on Brad (orange)
  • ✓ book held by Maria (purple)
  • ✓ book held by Aisha (orange)
  • ✓ book held by Brad (green)
  • ✓ no character's book matches their own hat
promptMaria wore the green hat, Aisha wore the purple hat, and Brad wore the orange hat -- and each of them carried a book in somebody else's colour: Maria's book was purple, Aisha's was orange, and Brad's was green. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t5-empty-hands
t5-empty-hands · r1gpt-image-2 · tier 5
composition2unprompted text!
  • ⇆ brad -- tan jacket, black trousers, short dark wavy hair Man is present but wears a green henley and blue jeans, and has short blond hair -- attributes wrong.
  • ⇆ Brad stands beside the table with both arms down at his sides He stands beside the table but arms are extended outward/forward with hands open, not hanging at sides.
  • ✓ wooden table (count 1)
  • ✓ Brad stands beside the table, both hands visible in frame
  • ⇆ both of Brad's hands are visibly empty, open, and hanging at his sides Hands are empty and open, but not hanging at sides; right hand rests on the tabletop, touching the surface.
  • ⇆ tabletop is completely bare No objects on tabletop, but Brad's right hand is resting on it, touching the surface.
promptBrad stood beside the wooden table with both hands empty and open at his sides, and the tabletop beside him was bare -- nothing on it at all, not a cup, not a book. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t5-empty-hands
t5-empty-hands · r2gpt-image-2 · tier 5
composition2unprompted text!
  • ⇆ brad -- tan jacket, black trousers, short dark wavy hair Man is present but wears a blue t-shirt and blue jeans, and has short blond wavy hair -- attributes wrong.
  • ⇆ Brad stands beside the table with both arms down at his sides He stands beside the table but arms are extended outward with palms up in a shrug, not down at his sides.
  • ✓ wooden table (count 1)
  • ✓ Brad stands beside the table, both hands visible in frame
  • ⇆ both of Brad's hands are visibly empty, open, and hanging at his sides Hands are empty and open, but not hanging at his sides -- they are extended outward in a shrug.
  • ✓ tabletop is completely bare
promptBrad stood beside the wooden table with both hands empty and open at his sides, and the tabletop beside him was bare -- nothing on it at all, not a cup, not a book. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t5-left-hand-lantern
t5-left-hand-lantern · r1gpt-image-2 · tier 5
composition3
  • ⇆ aisha -- green dress, gold hoop earrings, long black hair Aisha has long black hair but wears a purple/magenta kurta with orange dupatta, not a green dress; earrings are gold teardrop shapes, not hoops.
  • ✓ mr-penguin -- orange beak, white belly, black back
  • ✓ Aisha raises the lantern high in her own LEFT hand
  • ✓ Aisha's right hand hangs straight down at her side, empty
  • ✓ Mr Penguin steps into frame from the right edge
  • ✓ lit lantern (count 1)
  • ✓ lantern in Aisha's own left hand (viewer's right side)
  • ✓ Mr Penguin only partly visible, cropped by right edge, roughly half out of frame
  • ✓ Aisha's right hand visibly empty; no second lantern or object
promptAisha raised the lantern high in her own left hand -- the hand on the viewer's right side of her body -- while her right hand hung straight down at her side, and Mr Penguin was only half in the picture, stepping in from the right edge of the frame. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t5-left-hand-lantern
t5-left-hand-lantern · r2gpt-image-2 · tier 5
composition2unprompted text!
  • ⇆ aisha -- green dress, gold hoop earrings, long black hair Aisha is present with black hair but wears a purple tunic, not green, and no visible gold hoop earrings.
  • ✓ mr-penguin -- orange beak, white belly, black back
  • ⇆ Aisha raises the lantern high in her own LEFT hand The raised arm is on the viewer's left side of her body, which is her own RIGHT hand, not left.
  • ⇆ Aisha lets her right hand hang straight down at her side, empty Her right hand is actually the one raised holding the lantern; the hanging empty hand is her left.
  • ✓ Mr Penguin steps into frame from the right edge
  • ✓ lit lantern (count 1)
  • ⇆ lantern is in Aisha's OWN left hand (viewer's right of her body), not right hand Lantern appears on viewer's LEFT side of her body, meaning her own right hand -- inverted from the requirement.
  • ✓ Mr Penguin only PARTLY visible, cropped by right edge, ~half out of picture
  • ✗ Aisha's right hand visibly empty at her side; no second lantern or anything held Her right hand is not empty at her side -- it is the raised hand holding the lantern.
promptAisha raised the lantern high in her own left hand -- the hand on the viewer's right side of her body -- while her right hand hung straight down at her side, and Mr Penguin was only half in the picture, stepping in from the right edge of the frame. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t5-seven-acorns
t5-seven-acorns · r1gpt-image-2 · tier 5
composition2
  • ✓ mr-penguin -- black back, white belly, orange beak
  • ✓ Mr Penguin lines the acorns up along the log
  • ⇆ acorns in the row (count 7) Enumeration shows 8 acorns, not 7.
  • ✓ acorn painted bright gold (count 1)
  • ⇆ plain brown acorns in the row (count 6) Enumeration shows 7 brown acorns, not 6.
  • ✓ fallen log (count 1)
  • ⇆ all seven acorns in ONE straight row along the log The acorns are in one straight row, but there are 8 not 7.
  • ⇆ gold acorn is FOURTH from the left -- three brown to left, three to right Gold is 4th from left with 3 brown to its left, but 4 brown to its right, not 3.
promptMr Penguin lined up exactly seven acorns in a single row along the fallen log, and the fourth acorn from the left was painted bright gold while the other six stayed plain brown. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t5-seven-acorns
t5-seven-acorns · r2gpt-image-2 · tier 5
composition2
  • ✓ mr-penguin -- black back, white belly, orange beak
  • ✓ Mr Penguin lines the acorns up along the log
  • ⇆ acorns in the row (count 7) Enumeration shows 8 acorns, not 7.
  • ✓ acorn in the row painted bright gold (count 1)
  • ⇆ plain brown acorns in the row (count 6) Enumeration shows 7 brown acorns, not 6.
  • ✓ fallen log (count 1)
  • ⇆ all seven acorns sit in ONE straight row along the log They are in one row, but there are 8, not 7.
  • ⇆ the gold acorn is the FOURTH from the left -- three brown to left, three to right Gold is 4th from left with 3 brown to left, but 4 brown to its right, not 3.
promptMr Penguin lined up exactly seven acorns in a single row along the fallen log, and the fourth acorn from the left was painted bright gold while the other six stayed plain brown. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t5-size-order
t5-size-order · r1gpt-image-2 · tier 5
composition4unprompted text!
  • ⇆ aisha -- green dress, gold hoop earrings, long black hair Girl wears a purple hijab and purple dress, no visible gold hoop earrings; attributes do not match.
  • ✓ Aisha arranges the three books along the shelf
  • ✓ books standing on the shelf (count 3)
  • ✓ largest book, blue
  • ✓ middle-sized book, green
  • ✓ smallest book, red
  • ✓ shelf (count 1)
  • ✓ books descend in size from left to right
  • ✓ blue leftmost, green middle, red rightmost
promptAisha lined three books along the shelf from biggest to smallest: the big blue one on the left, a medium green one in the middle, and the small red one on the right. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.
t5-size-order
t5-size-order · r2gpt-image-2 · tier 5
composition4unprompted text!
  • ⇆ aisha -- green dress, gold hoop earrings, long black hair Has gold hoop earring and black hair (in ponytail), but wearing a purple hoodie, not a green dress.
  • ✓ Aisha arranges the three books along the shelf
  • ✓ books standing on the shelf (count 3)
  • ✓ largest book, blue
  • ✓ middle-sized book, green
  • ✓ smallest book, red
  • ✓ shelf
  • ✓ books descend in size from left to right
  • ✓ blue leftmost, green middle, red rightmost
promptAisha lined three books along the shelf from biggest to smallest: the big blue one on the left, a medium green one in the middle, and the small red one on the right. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded.

gpt-image-2 · arm B -- structured beat prompt

$0.040 / imageelement recall91%misbound9%composition4.3joint-correct8/10unprompted text0
t5-crossed-hats-trio
t5-crossed-hats-trio · r1gpt-image-2 · tier 5
composition5
  • ✓ maria -- green hat, orange thick-rimmed glasses, grey jacket
  • ✓ aisha -- purple hat, gold hoop earrings, green dress
  • ✓ brad -- orange hat, tan jacket, black trousers
  • ✓ three stand in a row facing viewer, each holding one book
  • ✓ hat on Maria (green)
  • ✓ hat on Aisha (purple)
  • ✓ hat on Brad (orange)
  • ✓ book held by Maria (purple)
  • ✓ book held by Aisha (orange)
  • ✓ book held by Brad (green)
  • ✓ no character's book matches their hat
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (green hat, orange thick-rimmed glasses, grey jacket) and Aisha (purple hat, gold hoop earrings, green dress) and Brad (orange hat, tan jacket, black trousers). Action: Maria, Aisha, and Brad stand in a row facing the viewer, each holding one book. Objects: exactly one green hat on Maria, exactly one purple hat on Aisha, exactly one orange hat on Brad, exactly one purple book held by Maria, exactly one orange book held by Aisha, exactly one green book held by Brad. Placement: NO character's book matches their own hat: every one of the three carries a book in a different colour from the hat on their head. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t5-crossed-hats-trio
t5-crossed-hats-trio · r2gpt-image-2 · tier 5
composition5
  • ✓ maria -- green hat, orange thick-rimmed glasses, grey jacket
  • ✓ aisha -- purple hat, gold hoop earrings, green dress
  • ✓ brad -- orange hat, tan jacket, black trousers
  • ✓ three stand in a row facing viewer, each holding one book
  • ✓ hat on Maria (green)
  • ✓ hat on Aisha (purple)
  • ✓ hat on Brad (orange)
  • ✓ book held by Maria (purple)
  • ✓ book held by Aisha (orange)
  • ✓ book held by Brad (green)
  • ✓ no character's book matches their hat
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (green hat, orange thick-rimmed glasses, grey jacket) and Aisha (purple hat, gold hoop earrings, green dress) and Brad (orange hat, tan jacket, black trousers). Action: Maria, Aisha, and Brad stand in a row facing the viewer, each holding one book. Objects: exactly one green hat on Maria, exactly one purple hat on Aisha, exactly one orange hat on Brad, exactly one purple book held by Maria, exactly one orange book held by Aisha, exactly one green book held by Brad. Placement: NO character's book matches their own hat: every one of the three carries a book in a different colour from the hat on their head. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t5-empty-hands
t5-empty-hands · r1gpt-image-2 · tier 5
composition5
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad stands beside the table with both arms down at his sides
  • ✓ wooden table (count 1)
  • ✓ Brad stands beside the table, both hands visible in frame
  • ✓ both of Brad's hands are visibly empty, open, and hanging at his sides
  • ✓ the tabletop is completely bare
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers, short dark wavy hair). Action: Brad stands beside the table with both arms down at his sides. Objects: exactly one wooden table. Placement: Brad stands beside the table, both hands visible in frame. Empty: both of Brad's hands are visibly empty, open, and hanging at his sides; the tabletop is completely bare, an empty wooden surface. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering. Do not show: any object held in, resting on, or touching either of Brad's hands; any object at all on the tabletop -- no cup, book, plant, tool, or clutter.
t5-empty-hands
t5-empty-hands · r2gpt-image-2 · tier 5
composition5
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad stands beside the table with both arms down at his sides
  • ✓ wooden table (count 1)
  • ✓ Brad stands beside the table, both hands visible in frame
  • ✓ both of Brad's hands empty and hanging at sides
  • ✓ tabletop completely bare
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers, short dark wavy hair). Action: Brad stands beside the table with both arms down at his sides. Objects: exactly one wooden table. Placement: Brad stands beside the table, both hands visible in frame. Empty: both of Brad's hands are visibly empty, open, and hanging at his sides; the tabletop is completely bare, an empty wooden surface. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering. Do not show: any object held in, resting on, or touching either of Brad's hands; any object at all on the tabletop -- no cup, book, plant, tool, or clutter.
t5-left-hand-lantern
t5-left-hand-lantern · r1gpt-image-2 · tier 5
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ mr-penguin -- orange beak, white belly, black back
  • ✓ Aisha raises the lantern high in her own LEFT hand
  • ✓ Aisha lets her right hand hang straight down at her side, empty
  • ✓ Mr Penguin steps into frame from the right edge
  • ✓ lit lantern (count 1)
  • ✓ lantern is in Aisha's OWN left hand (viewer's right side of her body)
  • ✓ Mr Penguin only PARTLY visible, cropped by right edge, roughly half out
  • ✓ Aisha's right hand visibly empty, nothing held
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair) and Mr Penguin (orange beak, white belly, black back). Action: Aisha raises the lantern high in her own LEFT hand; Aisha lets her right hand hang straight down at her side, empty; Mr Penguin steps into frame from the right edge. Objects: exactly one lit lantern. Placement: the lantern is in Aisha's OWN left hand -- appearing on the viewer's right side of her body -- and not in her right hand; Mr Penguin is only PARTLY visible, cropped by the right edge of the frame, roughly half his body out of the picture. Empty: Aisha's right hand is visibly empty, hanging down at her side. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering. Do not show: any second lantern, or anything at all held in Aisha's right hand.
t5-left-hand-lantern
t5-left-hand-lantern · r2gpt-image-2 · tier 5
composition2
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ mr-penguin -- orange beak, white belly, black back
  • ⇆ Aisha raises the lantern high in her own LEFT hand The raised arm is on the viewer's right side of Aisha, which is her own RIGHT arm, not her left.
  • ⇆ Aisha lets her right hand hang straight down at her side, empty The empty hanging hand is on viewer's left, which is Aisha's LEFT hand, not right.
  • ✓ Mr Penguin steps into frame from the right edge
  • ✓ lit lantern (count 1)
  • ⇆ lantern in Aisha's own left hand, viewer's right side Lantern is on viewer's right side of body but that is her own right hand, not left.
  • ✓ Mr Penguin only partly visible, cropped by right edge, ~half out
  • ⇆ Aisha's right hand visibly empty, hanging down; no second lantern One hand hangs empty and no second lantern appears, but the empty hand is her left, not her right as required.
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair) and Mr Penguin (orange beak, white belly, black back). Action: Aisha raises the lantern high in her own LEFT hand; Aisha lets her right hand hang straight down at her side, empty; Mr Penguin steps into frame from the right edge. Objects: exactly one lit lantern. Placement: the lantern is in Aisha's OWN left hand -- appearing on the viewer's right side of her body -- and not in her right hand; Mr Penguin is only PARTLY visible, cropped by the right edge of the frame, roughly half his body out of the picture. Empty: Aisha's right hand is visibly empty, hanging down at her side. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering. Do not show: any second lantern, or anything at all held in Aisha's right hand.
t5-seven-acorns
t5-seven-acorns · r1gpt-image-2 · tier 5
composition5
  • ✓ mr-penguin -- black back, white belly, orange beak
  • ✓ Mr Penguin lines the acorns up along the log
  • ✓ 7 acorns in row
  • ✓ 1 gold acorn in row
  • ✓ 6 plain brown acorns
  • ✓ 1 fallen log
  • ✓ all seven acorns in ONE straight row along the log
  • ✓ gold acorn is FOURTH from left, 3 brown left, 3 brown right
promptScene: a simple bright classroom setting, plain background. Subjects: Mr Penguin (black back, white belly, orange beak). Action: Mr Penguin lines the acorns up along the log. Objects: exactly seven acorns in the row, exactly one acorn in the row painted bright gold, exactly six plain brown acorns in the row, exactly one fallen log. Placement: all seven acorns sit in ONE straight row along the log, none stacked or scattered; the gold acorn is the FOURTH from the left end of the row -- three brown acorns to its left, three to its right. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t5-seven-acorns
t5-seven-acorns · r2gpt-image-2 · tier 5
composition2
  • ✓ mr-penguin -- black back, white belly, orange beak
  • ✓ Mr Penguin lines the acorns up along the log
  • ⇆ acorns in the row (count 7) I counted 8 acorns, not 7.
  • ✓ one gold acorn in the row
  • ⇆ plain brown acorns (count 6) I counted 7 brown acorns, not 6.
  • ✓ fallen log (count 1)
  • ⇆ all seven acorns in one straight row along the log Acorns are in one straight row, but total is 8 not 7.
  • ⇆ gold acorn is FOURTH from the left, three brown left and three right Gold is 4th from left with 3 brown to its left, but 4 brown to its right, not 3.
promptScene: a simple bright classroom setting, plain background. Subjects: Mr Penguin (black back, white belly, orange beak). Action: Mr Penguin lines the acorns up along the log. Objects: exactly seven acorns in the row, exactly one acorn in the row painted bright gold, exactly six plain brown acorns in the row, exactly one fallen log. Placement: all seven acorns sit in ONE straight row along the log, none stacked or scattered; the gold acorn is the FOURTH from the left end of the row -- three brown acorns to its left, three to its right. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t5-size-order
t5-size-order · r1gpt-image-2 · tier 5
composition4
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha arranges the three books along the shelf
  • ✓ three books standing on the shelf
  • ✓ largest book (blue)
  • ✓ middle-sized book (green)
  • ✓ smallest book (red)
  • ✓ shelf
  • ✓ books descend in size from left to right
  • ✓ blue leftmost, green middle, red rightmost
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair). Action: Aisha arranges the three books along the shelf. Objects: exactly three books standing on the shelf, exactly one blue book that is the LARGEST of the three, exactly one green book that is MIDDLE-sized, exactly one red book that is the SMALLEST of the three, exactly one shelf. Placement: the three books descend in size from left to right: biggest at the left end, smallest at the right end; colour and position agree exactly -- blue leftmost, green in the middle, red rightmost. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t5-size-order
t5-size-order · r2gpt-image-2 · tier 5
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha arranges the three books along the shelf
  • ✓ books standing on the shelf (count 3)
  • ✓ largest book, blue
  • ✓ middle-sized book, green
  • ✓ smallest book, red
  • ✓ shelf (count 1)
  • ✓ books descend in size left to right
  • ✓ blue leftmost, green middle, red rightmost
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair). Action: Aisha arranges the three books along the shelf. Objects: exactly three books standing on the shelf, exactly one blue book that is the LARGEST of the three, exactly one green book that is MIDDLE-sized, exactly one red book that is the SMALLEST of the three, exactly one shelf. Placement: the three books descend in size from left to right: biggest at the left end, smallest at the right end; colour and position agree exactly -- blue leftmost, green in the middle, red rightmost. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.

gpt-image-2 · arm C -- structured + references

$0.040 / imageelement recall84%misbound16%composition4.1joint-correct7/10unprompted text0
t5-crossed-hats-trio
t5-crossed-hats-trio · r1gpt-image-2 · tier 5 +ref
composition5
  • ✓ maria -- green hat, orange thick-rimmed glasses, grey jacket
  • ✓ aisha -- purple hat, gold hoop earrings, green dress
  • ✓ brad -- orange hat, tan jacket, black trousers
  • ✓ Maria, Aisha, and Brad stand in a row facing the viewer, each holding one book
  • ✓ hat on Maria (count 1; green)
  • ✓ hat on Aisha (count 1; purple)
  • ✓ hat on Brad (count 1; orange)
  • ✓ book held by Maria (count 1; purple)
  • ✓ book held by Aisha (count 1; orange)
  • ✓ book held by Brad (count 1; green)
  • ✓ No character's book matches their own hat
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (green hat, orange thick-rimmed glasses, grey jacket) and Aisha (purple hat, gold hoop earrings, green dress) and Brad (orange hat, tan jacket, black trousers). Action: Maria, Aisha, and Brad stand in a row facing the viewer, each holding one book. Objects: exactly one green hat on Maria, exactly one purple hat on Aisha, exactly one orange hat on Brad, exactly one purple book held by Maria, exactly one orange book held by Aisha, exactly one green book held by Brad. Placement: NO character's book matches their own hat: every one of the three carries a book in a different colour from the hat on their head. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t5-crossed-hats-trio
t5-crossed-hats-trio · r2gpt-image-2 · tier 5 +ref
composition5
  • ✓ maria -- green hat, orange thick-rimmed glasses, grey jacket
  • ✓ aisha -- purple hat, gold hoop earrings, green dress
  • ✓ brad -- orange hat, tan jacket, black trousers
  • ✓ three stand in a row facing viewer, each holding one book
  • ✓ hat on Maria (green)
  • ✓ hat on Aisha (purple)
  • ✓ hat on Brad (orange)
  • ✓ book held by Maria (purple)
  • ✓ book held by Aisha (orange)
  • ✓ book held by Brad (green)
  • ✓ no character's book matches their own hat
promptScene: a simple bright classroom setting, plain background. Subjects: Maria (green hat, orange thick-rimmed glasses, grey jacket) and Aisha (purple hat, gold hoop earrings, green dress) and Brad (orange hat, tan jacket, black trousers). Action: Maria, Aisha, and Brad stand in a row facing the viewer, each holding one book. Objects: exactly one green hat on Maria, exactly one purple hat on Aisha, exactly one orange hat on Brad, exactly one purple book held by Maria, exactly one orange book held by Aisha, exactly one green book held by Brad. Placement: NO character's book matches their own hat: every one of the three carries a book in a different colour from the hat on their head. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t5-empty-hands
t5-empty-hands · r1gpt-image-2 · tier 5 +ref
composition5
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad stands beside the table with both arms down at his sides
  • ✓ wooden table (count 1)
  • ✓ Brad stands beside the table, both hands visible in frame
  • ✓ both of Brad's hands are visibly empty and hanging at sides
  • ✓ tabletop is completely bare
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers, short dark wavy hair). Action: Brad stands beside the table with both arms down at his sides. Objects: exactly one wooden table. Placement: Brad stands beside the table, both hands visible in frame. Empty: both of Brad's hands are visibly empty, open, and hanging at his sides; the tabletop is completely bare, an empty wooden surface. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering. Do not show: any object held in, resting on, or touching either of Brad's hands; any object at all on the tabletop -- no cup, book, plant, tool, or clutter.
t5-empty-hands
t5-empty-hands · r2gpt-image-2 · tier 5 +ref
composition5
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad stands beside the table with both arms down at his sides
  • ✓ wooden table (count 1)
  • ✓ Brad stands beside the table, both hands visible in frame
  • ✓ both of Brad's hands are visibly empty
  • ✓ the tabletop is completely bare
promptScene: a simple bright classroom setting, plain background. Subjects: Brad (tan jacket, black trousers, short dark wavy hair). Action: Brad stands beside the table with both arms down at his sides. Objects: exactly one wooden table. Placement: Brad stands beside the table, both hands visible in frame. Empty: both of Brad's hands are visibly empty, open, and hanging at his sides; the tabletop is completely bare, an empty wooden surface. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering. Do not show: any object held in, resting on, or touching either of Brad's hands; any object at all on the tabletop -- no cup, book, plant, tool, or clutter.
t5-left-hand-lantern
t5-left-hand-lantern · r1gpt-image-2 · tier 5 +ref
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ mr-penguin -- orange beak, white belly, black back
  • ✓ Aisha raises lantern high in her own LEFT hand
  • ✓ right hand hangs straight down empty
  • ✓ Mr Penguin steps into frame from right edge
  • ✓ lit lantern (count 1)
  • ✓ lantern in Aisha's own left hand (viewer's right side of body)
  • ✓ Mr Penguin partly visible, cropped by right edge, ~half out
  • ✓ right hand empty; no second lantern or object
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair) and Mr Penguin (orange beak, white belly, black back). Action: Aisha raises the lantern high in her own LEFT hand; Aisha lets her right hand hang straight down at her side, empty; Mr Penguin steps into frame from the right edge. Objects: exactly one lit lantern. Placement: the lantern is in Aisha's OWN left hand -- appearing on the viewer's right side of her body -- and not in her right hand; Mr Penguin is only PARTLY visible, cropped by the right edge of the frame, roughly half his body out of the picture. Empty: Aisha's right hand is visibly empty, hanging down at her side. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering. Do not show: any second lantern, or anything at all held in Aisha's right hand.
t5-left-hand-lantern
t5-left-hand-lantern · r2gpt-image-2 · tier 5 +ref
composition2
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ mr-penguin -- orange beak, white belly, black back
  • ⇆ Aisha raises the lantern high in her own LEFT hand Lantern is raised, but held in her own RIGHT hand (viewer's left side of her body), not her left.
  • ⇆ Aisha lets her right hand hang straight down at her side, empty She has an empty hand hanging down, but it is her LEFT hand, not her right; handedness reversed.
  • ✓ Mr Penguin steps into frame from the right edge
  • ✓ lit lantern (count 1)
  • ⇆ lantern is in Aisha's OWN left hand (viewer's right side of her body), not right hand Lantern appears on viewer's LEFT side of her body, held in her own right hand -- opposite of required.
  • ⇆ Mr Penguin only PARTLY visible, cropped by right edge, roughly half body out Penguin is fairly fully visible; only a small sliver appears cropped, not roughly half.
  • ⇆ Aisha's right hand is visibly empty, hanging at her side -- no second lantern or anything held An empty hand hangs at her side and no second lantern exists, but the empty hand is her LEFT, not her right; her right hand holds the lantern.
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair) and Mr Penguin (orange beak, white belly, black back). Action: Aisha raises the lantern high in her own LEFT hand; Aisha lets her right hand hang straight down at her side, empty; Mr Penguin steps into frame from the right edge. Objects: exactly one lit lantern. Placement: the lantern is in Aisha's OWN left hand -- appearing on the viewer's right side of her body -- and not in her right hand; Mr Penguin is only PARTLY visible, cropped by the right edge of the frame, roughly half his body out of the picture. Empty: Aisha's right hand is visibly empty, hanging down at her side. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering. Do not show: any second lantern, or anything at all held in Aisha's right hand.
t5-seven-acorns
t5-seven-acorns · r1gpt-image-2 · tier 5 +ref
composition2
  • ✓ mr-penguin -- black back, white belly, orange beak
  • ✓ Mr Penguin lines the acorns up along the log
  • ⇆ acorns in the row (count 7) Enumeration shows 8 acorns, not 7.
  • ✓ one gold acorn in the row
  • ⇆ plain brown acorns (count 6) Enumeration shows 7 brown acorns, not 6.
  • ✓ fallen log (count 1)
  • ⇆ all seven acorns in one straight row, none stacked or scattered Row is straight but contains 8 acorns, not 7.
  • ⇆ gold acorn is FOURTH from the left, three brown left/three brown right Gold acorn is 4th from left but has 3 browns left and 4 browns right.
promptScene: a simple bright classroom setting, plain background. Subjects: Mr Penguin (black back, white belly, orange beak). Action: Mr Penguin lines the acorns up along the log. Objects: exactly seven acorns in the row, exactly one acorn in the row painted bright gold, exactly six plain brown acorns in the row, exactly one fallen log. Placement: all seven acorns sit in ONE straight row along the log, none stacked or scattered; the gold acorn is the FOURTH from the left end of the row -- three brown acorns to its left, three to its right. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t5-seven-acorns
t5-seven-acorns · r2gpt-image-2 · tier 5 +ref
composition2
  • ✓ mr-penguin -- black back, white belly, orange beak
  • ⇆ Mr Penguin lines the acorns up along the log Acorns are lined up on the log but the penguin is not actively arranging them; just standing behind.
  • ⇆ acorns in the row (count 7) Enumeration shows 8 acorns, not 7.
  • ✓ gold acorn (count 1)
  • ⇆ plain brown acorns (count 6) Enumeration shows 7 brown acorns, not 6.
  • ✓ fallen log (count 1)
  • ⇆ all seven acorns in ONE straight row, none stacked or scattered Acorns are in one straight row but there are 8, not 7.
  • ⇆ gold acorn is FOURTH from left (three brown left, three right) Gold acorn is 4th from left, but has 3 brown to left and 4 brown to right.
promptScene: a simple bright classroom setting, plain background. Subjects: Mr Penguin (black back, white belly, orange beak). Action: Mr Penguin lines the acorns up along the log. Objects: exactly seven acorns in the row, exactly one acorn in the row painted bright gold, exactly six plain brown acorns in the row, exactly one fallen log. Placement: all seven acorns sit in ONE straight row along the log, none stacked or scattered; the gold acorn is the FOURTH from the left end of the row -- three brown acorns to its left, three to its right. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t5-size-order
t5-size-order · r1gpt-image-2 · tier 5 +ref
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha arranges the three books along the shelf
  • ✓ books standing on the shelf (count 3)
  • ✓ largest book, blue
  • ✓ middle-sized book, green
  • ✓ smallest book, red
  • ✓ shelf (count 1)
  • ✓ books descend in size from left to right
  • ✓ blue leftmost, green middle, red rightmost
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair). Action: Aisha arranges the three books along the shelf. Objects: exactly three books standing on the shelf, exactly one blue book that is the LARGEST of the three, exactly one green book that is MIDDLE-sized, exactly one red book that is the SMALLEST of the three, exactly one shelf. Placement: the three books descend in size from left to right: biggest at the left end, smallest at the right end; colour and position agree exactly -- blue leftmost, green in the middle, red rightmost. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.
t5-size-order
t5-size-order · r2gpt-image-2 · tier 5 +ref
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha arranges the three books along the shelf
  • ✓ books standing on the shelf (count 3)
  • ✓ largest book, blue
  • ✓ middle-sized book, green
  • ✓ smallest book, red
  • ✓ shelf (count 1)
  • ✓ books descend in size from left to right
  • ✓ colour-position: blue left, green middle, red right
promptScene: a simple bright classroom setting, plain background. Subjects: Aisha (green dress, gold hoop earrings, long black hair). Action: Aisha arranges the three books along the shelf. Objects: exactly three books standing on the shelf, exactly one blue book that is the LARGEST of the three, exactly one green book that is MIDDLE-sized, exactly one red book that is the SMALLEST of the three, exactly one shelf. Placement: the three books descend in size from left to right: biggest at the left end, smallest at the right end; colour and position agree exactly -- blue leftmost, green in the middle, red rightmost. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Do not add any other characters or extra props; keep every surface free of lettering.

nano-banana

nano-banana · arm B -- structured beat prompt

$0.039 / imageelement recall92%misbound7%composition4.4joint-correct7/10unprompted text0
t5-crossed-hats-trio
t5-crossed-hats-trio · r1nano-banana · tier 5
composition5
  • ✓ maria -- green hat, orange thick-rimmed glasses, grey jacket
  • ✓ aisha -- purple hat, gold hoop earrings, green dress
  • ✓ brad -- orange hat, tan jacket, black trousers
  • ✓ Maria, Aisha, and Brad stand in a row facing viewer, each holding one book
  • ✓ hat on Maria (green)
  • ✓ hat on Aisha (purple)
  • ✓ hat on Brad (orange)
  • ✓ book held by Maria (purple)
  • ✓ book held by Aisha (orange)
  • ✓ book held by Brad (green)
  • ✓ NO character's book matches their own hat
promptA graphic-novel illustration of Maria (green hat, orange thick-rimmed glasses, grey jacket) and Aisha (purple hat, gold hoop earrings, green dress) and Brad (orange hat, tan jacket, black trousers), Maria, Aisha, and Brad stand in a row facing the viewer, each holding one book, with exactly one green hat on Maria, exactly one purple hat on Aisha, exactly one orange hat on Brad, exactly one purple book held by Maria, exactly one orange book held by Aisha, exactly one green book held by Brad, NO character's book matches their own hat: every one of the three carries a book in a different colour from the hat on their head, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-crossed-hats-trio
t5-crossed-hats-trio · r2nano-banana · tier 5
composition5
  • ✓ maria -- green hat, orange thick-rimmed glasses, grey jacket
  • ✓ aisha -- purple hat, gold hoop earrings, green dress
  • ✓ brad -- orange hat, tan jacket, black trousers
  • ✓ three stand in a row facing viewer, each holding one book
  • ✓ hat on Maria (green)
  • ✓ hat on Aisha (purple)
  • ✓ hat on Brad (orange)
  • ✓ book held by Maria (purple)
  • ✓ book held by Aisha (orange)
  • ✓ book held by Brad (green)
  • ✓ no character's book matches their own hat
promptA graphic-novel illustration of Maria (green hat, orange thick-rimmed glasses, grey jacket) and Aisha (purple hat, gold hoop earrings, green dress) and Brad (orange hat, tan jacket, black trousers), Maria, Aisha, and Brad stand in a row facing the viewer, each holding one book, with exactly one green hat on Maria, exactly one purple hat on Aisha, exactly one orange hat on Brad, exactly one purple book held by Maria, exactly one orange book held by Aisha, exactly one green book held by Brad, NO character's book matches their own hat: every one of the three carries a book in a different colour from the hat on their head, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-empty-hands
t5-empty-hands · r1nano-banana · tier 5
composition5
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad stands beside the table with both arms down at his sides
  • ✓ wooden table (count 1)
  • ✓ Brad stands beside the table, both hands visible in frame
  • ✓ both of Brad's hands are visibly empty, open, and hanging at his sides
  • ✓ tabletop is completely bare
promptA graphic-novel illustration of Brad (tan jacket, black trousers, short dark wavy hair), Brad stands beside the table with both arms down at his sides, with exactly one wooden table, Brad stands beside the table, both hands visible in frame, both of Brad's hands are visibly empty, open, and hanging at his sides; the tabletop is completely bare, an empty wooden surface, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-empty-hands
t5-empty-hands · r2nano-banana · tier 5
composition4
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad stands beside the table with both arms down at his sides
  • ⇆ wooden table (count 1) Checklist asks for 1 table; image shows the main table plus additional student desks visible on the right.
  • ✓ Brad stands beside the table, both hands visible in frame
  • ✓ both of Brad's hands visibly empty and open at sides
  • ✓ tabletop completely bare
promptA graphic-novel illustration of Brad (tan jacket, black trousers, short dark wavy hair), Brad stands beside the table with both arms down at his sides, with exactly one wooden table, Brad stands beside the table, both hands visible in frame, both of Brad's hands are visibly empty, open, and hanging at his sides; the tabletop is completely bare, an empty wooden surface, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-left-hand-lantern
t5-left-hand-lantern · r1nano-banana · tier 5
composition3
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ mr-penguin -- orange beak, white belly, black back
  • ✓ Aisha raises the lantern high in her own LEFT hand
  • ✓ Aisha lets her right hand hang straight down at her side, empty
  • ⇆ Mr Penguin steps into frame from the right edge Penguin is on the right side but fully visible, not cropped/stepping in.
  • ✓ lit lantern (count 1; no special attributes)
  • ✓ lantern in Aisha's OWN left hand (viewer's right side of body), not in her right
  • ✗ Mr Penguin is only PARTLY visible, cropped by right edge, roughly half out of picture Penguin is fully within frame, not cropped.
  • ✓ Aisha's right hand visibly empty; no second lantern or anything held
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair) and Mr Penguin (orange beak, white belly, black back), Aisha raises the lantern high in her own LEFT hand; Aisha lets her right hand hang straight down at her side, empty; Mr Penguin steps into frame from the right edge, with exactly one lit lantern, the lantern is in Aisha's OWN left hand -- appearing on the viewer's right side of her body -- and not in her right hand; Mr Penguin is only PARTLY visible, cropped by the right edge of the frame, roughly half his body out of the picture, Aisha's right hand is visibly empty, hanging down at her side, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-left-hand-lantern
t5-left-hand-lantern · r2nano-banana · tier 5
composition2
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ mr-penguin -- orange beak, white belly, black back
  • ⇆ Aisha raises the lantern high in her own LEFT hand The raised arm is on the viewer's left side of her body, which corresponds to Aisha's OWN right hand/arm, not her left.
  • ⇆ Aisha lets her right hand hang straight down at her side, empty The hand hanging down is on viewer's right of her body, i.e., her own LEFT hand, not right.
  • ✓ Mr Penguin steps into frame from the right edge
  • ✓ lit lantern (count 1)
  • ⇆ lantern in Aisha's OWN left hand -- viewer's right side of her body Lantern appears on viewer's LEFT of her body, i.e., her own right hand -- opposite of requirement.
  • ✓ Mr Penguin partly visible, cropped by right edge, roughly half out
  • ⇆ Aisha's right hand visibly empty at her side One hand does hang down empty, but it is her LEFT hand, not her right; her right hand actually holds the lantern raised. Emptiness bound to wrong hand.
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair) and Mr Penguin (orange beak, white belly, black back), Aisha raises the lantern high in her own LEFT hand; Aisha lets her right hand hang straight down at her side, empty; Mr Penguin steps into frame from the right edge, with exactly one lit lantern, the lantern is in Aisha's OWN left hand -- appearing on the viewer's right side of her body -- and not in her right hand; Mr Penguin is only PARTLY visible, cropped by the right edge of the frame, roughly half his body out of the picture, Aisha's right hand is visibly empty, hanging down at her side, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-seven-acorns
t5-seven-acorns · r1nano-banana · tier 5
composition5
  • ✓ mr-penguin -- black back, white belly, orange beak
  • ✓ Mr Penguin lines the acorns up along the log
  • ✓ acorns in the row (count 7)
  • ✓ acorn in the row painted bright gold (count 1)
  • ✓ plain brown acorns in the row (count 6)
  • ✓ fallen log (count 1)
  • ✓ all seven acorns sit in ONE straight row along the log, none stacked or scattered
  • ✓ gold acorn is the FOURTH from the left -- three brown to its left, three to its right
promptA graphic-novel illustration of Mr Penguin (black back, white belly, orange beak), Mr Penguin lines the acorns up along the log, with exactly seven acorns in the row, exactly one acorn in the row painted bright gold, exactly six plain brown acorns in the row, exactly one fallen log, all seven acorns sit in ONE straight row along the log, none stacked or scattered; the gold acorn is the FOURTH from the left end of the row -- three brown acorns to its left, three to its right, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-seven-acorns
t5-seven-acorns · r2nano-banana · tier 5
composition5
  • ✓ mr-penguin -- attributes: black back, white belly, orange beak
  • ✓ Mr Penguin lines the acorns up along the log
  • ✓ acorns in the row (count 7)
  • ✓ acorn in the row painted bright gold (count 1)
  • ✓ plain brown acorns in the row (count 6)
  • ✓ fallen log (count 1)
  • ✓ all seven acorns sit in ONE straight row along the log, none stacked or scattered
  • ✓ gold acorn is the FOURTH from the left -- three brown to its left, three to its right
promptA graphic-novel illustration of Mr Penguin (black back, white belly, orange beak), Mr Penguin lines the acorns up along the log, with exactly seven acorns in the row, exactly one acorn in the row painted bright gold, exactly six plain brown acorns in the row, exactly one fallen log, all seven acorns sit in ONE straight row along the log, none stacked or scattered; the gold acorn is the FOURTH from the left end of the row -- three brown acorns to its left, three to its right, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-size-order
t5-size-order · r1nano-banana · tier 5
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha arranges the three books along the shelf
  • ✓ books standing on the shelf (count 3)
  • ✓ largest book, blue
  • ✓ middle-sized book, green
  • ✓ smallest book, red
  • ✓ shelf (count 1)
  • ✓ books descend in size from left to right
  • ✓ blue leftmost, green middle, red rightmost
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair), Aisha arranges the three books along the shelf, with exactly three books standing on the shelf, exactly one blue book that is the LARGEST of the three, exactly one green book that is MIDDLE-sized, exactly one red book that is the SMALLEST of the three, exactly one shelf, the three books descend in size from left to right: biggest at the left end, smallest at the right end; colour and position agree exactly -- blue leftmost, green in the middle, red rightmost, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-size-order
t5-size-order · r2nano-banana · tier 5
composition5
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha arranges the three books along the shelf
  • ✓ books standing on the shelf (count 3)
  • ✓ LARGEST book, blue
  • ✓ MIDDLE-sized book, green
  • ✓ SMALLEST book, red
  • ✓ shelf
  • ✓ books descend in size left to right
  • ✓ blue leftmost, green middle, red rightmost
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair), Aisha arranges the three books along the shelf, with exactly three books standing on the shelf, exactly one blue book that is the LARGEST of the three, exactly one green book that is MIDDLE-sized, exactly one red book that is the SMALLEST of the three, exactly one shelf, the three books descend in size from left to right: biggest at the left end, smallest at the right end; colour and position agree exactly -- blue leftmost, green in the middle, red rightmost, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.

nano-banana · arm C -- structured + references

$0.039 / imageelement recall77%misbound22%composition3.5joint-correct4/10unprompted text0
t5-crossed-hats-trio
t5-crossed-hats-trio · r1nano-banana · tier 5 +ref
composition5
  • ✓ maria -- green hat, orange thick-rimmed glasses, grey jacket
  • ✓ aisha -- purple hat, gold hoop earrings, green dress
  • ✓ brad -- orange hat, tan jacket, black trousers
  • ✓ three stand in a row facing viewer, each holding one book
  • ✓ hat on Maria (green)
  • ✓ hat on Aisha (purple)
  • ✓ hat on Brad (orange)
  • ✓ book held by Maria (purple)
  • ✓ book held by Aisha (orange)
  • ✓ book held by Brad (green)
  • ✓ no character's book matches their own hat
promptA graphic-novel illustration of Maria (green hat, orange thick-rimmed glasses, grey jacket) and Aisha (purple hat, gold hoop earrings, green dress) and Brad (orange hat, tan jacket, black trousers), Maria, Aisha, and Brad stand in a row facing the viewer, each holding one book, with exactly one green hat on Maria, exactly one purple hat on Aisha, exactly one orange hat on Brad, exactly one purple book held by Maria, exactly one orange book held by Aisha, exactly one green book held by Brad, NO character's book matches their own hat: every one of the three carries a book in a different colour from the hat on their head, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-crossed-hats-trio
t5-crossed-hats-trio · r2nano-banana · tier 5 +ref
composition5
  • ✓ maria -- green hat, orange thick-rimmed glasses, grey jacket
  • ✓ aisha -- purple hat, gold hoop earrings, green dress
  • ✓ brad -- orange hat, tan jacket, black trousers
  • ✓ three stand in a row facing viewer, each holding one book
  • ✓ hat on Maria (1; green)
  • ✓ hat on Aisha (1; purple)
  • ✓ hat on Brad (1; orange)
  • ✓ book held by Maria (1; purple)
  • ✓ book held by Aisha (1; orange)
  • ✓ book held by Brad (1; green)
  • ✓ no character's book matches their own hat
  • ✓
promptA graphic-novel illustration of Maria (green hat, orange thick-rimmed glasses, grey jacket) and Aisha (purple hat, gold hoop earrings, green dress) and Brad (orange hat, tan jacket, black trousers), Maria, Aisha, and Brad stand in a row facing the viewer, each holding one book, with exactly one green hat on Maria, exactly one purple hat on Aisha, exactly one orange hat on Brad, exactly one purple book held by Maria, exactly one orange book held by Aisha, exactly one green book held by Brad, NO character's book matches their own hat: every one of the three carries a book in a different colour from the hat on their head, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-empty-hands
t5-empty-hands · r1nano-banana · tier 5 +ref
composition3
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad stands beside the table with both arms down at his sides
  • ⇆ wooden table (count 1) Checklist asks for 1 table; multiple desks are visible in the classroom.
  • ✓ Brad stands beside the table, both hands visible in frame
  • ✓ both of Brad's hands are visibly empty, open, and hanging at his sides
  • ✓ the tabletop is completely bare
promptA graphic-novel illustration of Brad (tan jacket, black trousers, short dark wavy hair), Brad stands beside the table with both arms down at his sides, with exactly one wooden table, Brad stands beside the table, both hands visible in frame, both of Brad's hands are visibly empty, open, and hanging at his sides; the tabletop is completely bare, an empty wooden surface, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-empty-hands
t5-empty-hands · r2nano-banana · tier 5 +ref
composition5
  • ✓ brad -- tan jacket, black trousers, short dark wavy hair
  • ✓ Brad stands beside the table with both arms down at his sides
  • ✓ wooden table (count 1)
  • ✓ Brad stands beside the table, both hands visible in frame
  • ✓ both of Brad's hands are visibly empty
  • ✓ tabletop completely bare
promptA graphic-novel illustration of Brad (tan jacket, black trousers, short dark wavy hair), Brad stands beside the table with both arms down at his sides, with exactly one wooden table, Brad stands beside the table, both hands visible in frame, both of Brad's hands are visibly empty, open, and hanging at his sides; the tabletop is completely bare, an empty wooden surface, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-left-hand-lantern
t5-left-hand-lantern · r1nano-banana · tier 5 +ref
composition3
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ⇆ mr-penguin -- orange beak, white belly, black back Penguin has white belly and black back but beak appears dark/black, not orange.
  • ⇆ Aisha raises the lantern high in her own LEFT hand The lantern appears on the viewer's right side of her body, which corresponds to her own left hand -- but the arm shown reaching out looks anatomically like her right arm extended across; ambiguous handedness. Given conservative rule, misbound.
  • ✓ Aisha lets her right hand hang straight down at her side, empty
  • ✓ Mr Penguin steps into frame from the right edge
  • ✓ lit lantern (count 1)
  • ✓ lantern in Aisha's own left hand (viewer's right of her body), not right hand
  • ⇆ Mr Penguin only PARTLY visible, cropped by right edge, ~half out of picture Penguin is at right edge but appears mostly fully visible, only slightly cropped, not half out.
  • ✓ Aisha's right hand visibly empty, hanging down; no second lantern or held object
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair) and Mr Penguin (orange beak, white belly, black back), Aisha raises the lantern high in her own LEFT hand; Aisha lets her right hand hang straight down at her side, empty; Mr Penguin steps into frame from the right edge, with exactly one lit lantern, the lantern is in Aisha's OWN left hand -- appearing on the viewer's right side of her body -- and not in her right hand; Mr Penguin is only PARTLY visible, cropped by the right edge of the frame, roughly half his body out of the picture, Aisha's right hand is visibly empty, hanging down at her side, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-left-hand-lantern
t5-left-hand-lantern · r2nano-banana · tier 5 +ref
composition3
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ⇆ mr-penguin -- orange beak, white belly, black back Penguin has black beak, not orange; belly and back correct.
  • ⇆ Aisha raises the lantern high in her own LEFT hand The lantern is raised in Aisha's own RIGHT hand (viewer's left side of her body is her right; she holds it on the viewer's right, which is her left)... actually on inspection the raised arm is on the viewer's right, which is Aisha's own left hand. Correct.
  • ✓ Aisha lets her right hand hang straight down at her side, empty
  • ⇆ Mr Penguin steps into frame from the right edge Penguin is on the right side but mostly fully visible, not clearly stepping in mid-motion.
  • ✓ lit lantern (count 1)
  • ✓ lantern is in Aisha's OWN left hand (viewer's right of her body), not her right
  • ✗ Mr Penguin only PARTLY visible, cropped by right edge, ~half out of frame Penguin is almost fully inside the frame; only feet slightly cropped at bottom, not half-cropped by right edge.
  • ✓ Aisha's right hand is visibly empty, hanging down; no second lantern or object in right hand
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair) and Mr Penguin (orange beak, white belly, black back), Aisha raises the lantern high in her own LEFT hand; Aisha lets her right hand hang straight down at her side, empty; Mr Penguin steps into frame from the right edge, with exactly one lit lantern, the lantern is in Aisha's OWN left hand -- appearing on the viewer's right side of her body -- and not in her right hand; Mr Penguin is only PARTLY visible, cropped by the right edge of the frame, roughly half his body out of the picture, Aisha's right hand is visibly empty, hanging down at her side, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-seven-acorns
t5-seven-acorns · r1nano-banana · tier 5 +ref
composition2
  • ✓ mr-penguin -- black back, white belly, orange beak
  • ✓ Mr Penguin lines the acorns up along the log
  • ⇆ acorns in the row (count 7) I count 8 acorns in the row, not 7.
  • ✓ acorn in the row painted bright gold (count 1)
  • ⇆ plain brown acorns in the row (count 6) I count 7 brown acorns, not 6.
  • ✓ fallen log (count 1)
  • ⇆ all seven acorns sit in ONE straight row along the log They are in one row but the total is 8, not 7.
  • ⇆ gold acorn is FOURTH from the left -- three brown to its left, three to its right Gold acorn is fourth from left, but four brown acorns sit to its right, not three.
promptA graphic-novel illustration of Mr Penguin (black back, white belly, orange beak), Mr Penguin lines the acorns up along the log, with exactly seven acorns in the row, exactly one acorn in the row painted bright gold, exactly six plain brown acorns in the row, exactly one fallen log, all seven acorns sit in ONE straight row along the log, none stacked or scattered; the gold acorn is the FOURTH from the left end of the row -- three brown acorns to its left, three to its right, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-seven-acorns
t5-seven-acorns · r2nano-banana · tier 5 +ref
composition2
  • ✓ mr-penguin with black back, white belly, orange beak
  • ✓ Mr Penguin lines the acorns up along the log
  • ⇆ acorns in the row (count 7) Enumeration shows 9 acorns, not 7.
  • ✓ acorn in the row painted bright gold (count 1)
  • ⇆ plain brown acorns in the row (count 6) 8 brown acorns counted, not 6.
  • ✓ fallen log (count 1)
  • ⇆ all seven acorns sit in ONE straight row along the log Row is straight but count is 9, not 7.
  • ⇆ gold acorn is FOURTH from left with three brown to left and three to right Gold acorn is 5th from left (4 brown left, 4 brown right).
promptA graphic-novel illustration of Mr Penguin (black back, white belly, orange beak), Mr Penguin lines the acorns up along the log, with exactly seven acorns in the row, exactly one acorn in the row painted bright gold, exactly six plain brown acorns in the row, exactly one fallen log, all seven acorns sit in ONE straight row along the log, none stacked or scattered; the gold acorn is the FOURTH from the left end of the row -- three brown acorns to its left, three to its right, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-size-order
t5-size-order · r1nano-banana · tier 5 +ref
composition4
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha arranges the three books along the shelf
  • ✓ books standing on the shelf (count 3)
  • ✓ largest book, blue
  • ✓ middle-sized book, green
  • ✓ smallest book, red
  • ✓ shelf (count 1)
  • ✓ books descend in size from left to right
  • ✓ colour/position: blue left, green middle, red right
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair), Aisha arranges the three books along the shelf, with exactly three books standing on the shelf, exactly one blue book that is the LARGEST of the three, exactly one green book that is MIDDLE-sized, exactly one red book that is the SMALLEST of the three, exactly one shelf, the three books descend in size from left to right: biggest at the left end, smallest at the right end; colour and position agree exactly -- blue leftmost, green in the middle, red rightmost, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.
t5-size-order
t5-size-order · r2nano-banana · tier 5 +ref
composition3
  • ✓ aisha -- green dress, gold hoop earrings, long black hair
  • ✓ Aisha arranges the three books along the shelf
  • ✓ books standing on the shelf (count 3)
  • ⇆ LARGEST book, blue Blue book is present and is the tallest, but the size difference is modest; it does read as largest so acceptable.
  • ⇆ MIDDLE-sized book, green Green book present but appears roughly the same size as the red book, not clearly middle-sized.
  • ⇆ SMALLEST book, red Red book appears about the same height as green, not visibly smallest.
  • ✓ shelf (count 1)
  • ⇆ books descend in size left to right Blue is tallest on left, but green and red are near-equal so descent is not clear.
  • ✓ colour-position agreement: blue left, green middle, red right
promptA graphic-novel illustration of Aisha (green dress, gold hoop earrings, long black hair), Aisha arranges the three books along the shelf, with exactly three books standing on the shelf, exactly one blue book that is the LARGEST of the three, exactly one green book that is MIDDLE-sized, exactly one red book that is the SMALLEST of the three, exactly one shelf, the three books descend in size from left to right: biggest at the left end, smallest at the right end; colour and position agree exactly -- blue leftmost, green in the middle, red rightmost, set in a simple bright classroom setting, plain background. Rendered as bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. Every surface is blank and free of any lettering or signs.

Full written results

the complete documents behind this page -- method, every table, the reads, and the standing caveats

_Last updated: 2026-08-07_

Experiment 5 ablation -- which part of "structured prompting" is actually doing the work

Why this ran

Experiment 5 reported that structured beat prompts beat raw paragraph piping by 20-27 element-recall points. That number was real but confounded: arm B changed four things at once relative to arm A. It injected each character's physical attributes, imposed labeled segment structure, stated counts and spatial relations explicitly, and appended an exclusion clause.

That distinction decides an engineering workstream. Production already injects character attributes on every page -- the [A] / [B] verbatim-copy block puts a 9th grader Caucasian boy wearing crimson jacket into every prompt. If attribute injection is where the win lives, production has already banked it and there is nothing to build. If the labeled structure and explicit counts are doing the work, production needs a real structured-beat assembler.

Four rungs, each adding one component to the one before, on the tier-4 binding-trap beats:

  1. attrs -- the paragraph + character attribute injection only
  2. structure -- + labeled segment structure
  3. counts -- + explicit counts and spatial relations
  4. exclusion -- + the exclusion clause (byte-identical to experiment 5's arm B, pinned by test)

2 models x 4 rungs x 4 beats x 2 replicates = 64 renders, opus-judged.

Results: element recall

ModelRungRecallMisboundCompositionJoint-correct
gpt-image-2attrs100%0%5.008/8
gpt-image-2structure90%6%4.385/8
gpt-image-2counts100%0%5.008/8
gpt-image-2exclusion98%2%4.887/8
nano-bananaattrs87%10%4.123/8
nano-bananastructure81%15%3.752/8
nano-bananacounts92%6%4.756/8
nano-bananaexclusion88%12%4.504/8

The first rung is the whole effect. Attribute injection alone puts gpt-image-2 at 100% recall, zero misbinding, and 8/8 joint-correct -- matching what experiment 5's fully structured arm B achieved, with none of arm B's other three components present. nano-banana's attrs rung lands at 87%, inside the range its arm B reached.

Rungs 2 through 4 move within a few points in both directions and never above rung 1's ceiling. The sequence is not monotonic on either model -- recall dips at structure and recovers at counts -- which is itself the tell: increments that were doing real work would accumulate rather than wobble. At n=8 cells per rung these movements are noise around a flat line, not a ranking.

Results: unprompted lettering

The exclusion clause looked worthless above. It is not -- it is doing a different job.

Modelattrsstructurecountsexclusion
gpt-image-26/88/86/80/8
nano-banana1/83/80/80/8

gpt-image-2 volunteers lettering in 75-100% of cells on every rung that lacks the exclusion clause, and in zero cells once it is present. One sentence is a complete off-switch, replicating experiment 5's finding on a different beat set.

The read

The two components that matter each fix a different failure, and neither of the other two fixes anything measured here:

  • Attribute injection carries element recall and character correctness. It is the entire 20-27 point gap.
  • The exclusion clause carries lettering suppression. It contributes nothing to recall and everything to keeping invented text off the page.
  • Labeled structure and explicit counts did not earn their place on these beats. See the 2026-08-07 correction below before generalizing that to harder scenes.

What this changes for production

Do not build a structured-beat prompt assembler on this evidence alone. Production already injects character attributes on every page, which means it has already captured the component that produced the entire measured win on these beats. The workstream that experiment 5's headline number appeared to justify does not survive the ablation as stated.

Correction added 2026-08-07, after the tier-5 beats were judged. The sentence above is true for tier-4 beats and should not be generalized past them. On the tier-5 compound traps -- several constraints that must all hold in one image -- the full structured prompt still beats the bare paragraph 91% to 65% on gpt-image-2. That tier was never put through this ladder, so we cannot say whether its 26-point gap is also just attribute injection or whether labeled structure finally starts earning its keep once a scene gets dense enough. Decomposing tier 5 is roughly 64 renders and it is the measurement that settles whether the assembler is inert or merely unnecessary at today's page complexity. Until that runs, treat this as "not justified at the complexity we ship today" rather than "never worth building."

Do add a text-suppression clause. Production's image-prompt builder has no exclusion clause of any kind -- a search of generateImagePromptsForParagraphs.ts for lettering, text, signage, or caption directives returns nothing. On the evidence above that is a 75-100% lettering rate on the lead model, fixable by appending one sentence to a prompt template. No other correctness win in this program costs so little: no pipeline change, no new component, no added inference cost.

What is worth measuring next is whether production's existing attribute injection is as good as the harness's. The harness injects a compact per-beat descriptor (Brad (tan jacket, black trousers, short dark wavy hair)); production copies a longer descriptor built from database fields and currently ships two known defects into it (a missing null guard that can emit a null-year-old ... human, and label markup leaking into the final prompt). Those are filed and patched. Until they land, an A-to-B comparison against live production prompts measures the defects as much as the design.

Standing caveats

n=8 cells per rung, one style, tier-4 beats only -- and that last limit turned out to matter more than expected once tier 5 was judged. The rungs are cumulative rather than factorial, so this isolates where the gain first appears but cannot detect an interaction between structure and counts. The attrs rung minus its injected attribute text reproduces experiment 5's arm A exactly, and the exclusion rung is byte-identical to its arm B -- both anchored by committed fixtures, so this ladder is directly comparable to the experiment 5 numbers rather than merely similar to them.

Cost ~$2.50.

_Last updated: 2026-08-04_

Harder beats for a saturated rubric (tier 5)

Experiment 5 measured scene accuracy and produced a real finding: replacing the story

paragraph with a structured beat prompt takes element recall from 74% to 100% on

gpt-image-2 and from 72% to 90% on nano-banana, with actor recall going from 24%

and 18% to 100% on both. It also produced a problem that blocks every follow-up.

**All 32 of gpt-image-2's structured cells scored exactly 100% element recall and

exactly 5.00 composition.** Tier 1 and tier 4 alike, binding traps included. A

metric pinned at the top of its range cannot separate good from excellent, cannot

catch a regression, and cannot say how much headroom is left. Until the beats got

harder, no further number about the lead model meant anything.

This document is the fix: a fifth tier of compound traps, plus the probe that

validated it. It also reports a second finding that came out of the probe and

matters more than the tier itself, because it constrains what the program can

measure at all.

What tier 5 is

Tier 4 scored atoms that mostly stood alone: a red scarf on Maria, three pots on

a shelf, a cat under a table. Each element could be satisfied without reference

to the others, which is why a strong model could sweep them. Tier 5 scores atoms

that constrain each other, so a model can hit every individual noun and still

fail the beat.

The prior being tested is the GenEval 2 prompt-level cliff: 84.4% correctness

when atoms are scored individually against 31.0% when the whole prompt has to be

right at once. Tier 4 measured the first number. Tier 5 is built to measure the

second.

BeatElementsFailure family it targets
t5-seven-acorns8exact count above the reliable shelf (seven) crossed with ordinal position ("the fourth from the left")
t5-empty-hands6required absence, the documented worst case, since these models have no mechanism for "no"
t5-crossed-hats-trio11three-way crossed attribute permutation, anti-correlated on purpose so nobody's book matches their own hat
t5-size-order9comparative size crossed with left-to-right ordering and color, all three of which must agree
t5-left-hand-lantern9handedness, disambiguated against the viewer, plus deliberate partial-frame cropping

Two mechanical additions came with the tier.

A new absence checklist category. Its entries carry a positive claim about

what the emptiness should look like plus the thing that must not appear. The GPT

family gets both, because exclusion clauses are endorsed there; Gemini gets the

claim only, because its playbook is explicit that negative prompting hurts on

that family. That split is a deliberate per-family difference, not an oversight.

A prompt-level metric, jointCorrect. A cell passes only if every element

landed. One wrong atom takes the cell from 1 to 0, which is why it stays

sensitive after per-element recall has already pinned at the top. It is computed

from the element list rather than asked of the judge, so it costs nothing and

cannot drift.

The tier-5 composition anchor was also tightened: a 5 now has to survive an art

director with no notes, and an image where every element is correct but the

anatomy is awkward or the focal point is muddled takes a 4. Tiers 1 and 4 keep

the original wording so their experiment 5 scores stay comparable.

Adding tier 5 changed no tier-1 or tier-4 prompt. All 133 prompts recorded in the

experiment 5 run directories rebuild byte-identically from the current code, so round

5 remains reproducible.

The probe

Five cells: every tier-5 beat once, on gpt-image-2, arm B (the structured prompt),

one replicate. This was validation of the instrument, not a measurement round.

The test-account token was expired, so the judge ran through the Anthropic key rather

than Bedrock. Same model, claude-opus-4-7, different route.

BeatRecallJointComposition
t5-empty-hands6/6 (100%)pass5
t5-crossed-hats-trio11/11 (100%)pass5
t5-size-order9/9 (100%)pass5
t5-left-hand-lantern8/9 (89%)fail4
t5-seven-acorns4/8 (50%)fail2
pooled38/43 (88%)3/5 (60%)4.2, spanning 2 to 5

Against experiment 5's 100% / 100% / 5.00 on the same model and the same arm. The

ceiling is broken and the scale has usable room under the top again.

What actually broke, and what did not

Deliberate cropping failed. Asked for Mr Penguin half out of frame at the

right edge, gpt-image-2 cropped a sliver instead. The model resists composing a

subject partly outside the picture, which is a real production constraint: story

pages often want a character entering or exiting a scene.

Counting failed, by one. Asked for seven acorns, the model rendered eight.

The ordinal it got right, which is the more interesting half: the gold acorn

landed fourth from the left exactly as asked, and the count around it drifted.

Three-way attribute permutation passed cleanly. Nine bindings across three

characters, deliberately anti-correlated, all correct. Tier 4's crossed-attribute

family is solved on this model, and scaling it from two-by-two to

three-by-three does not reopen it. Report that as a capability, not a gap.

Comparative size with ordering passed. Biggest blue on the left through

smallest red on the right, correct on every axis. The one composition point it

lost was an anatomy problem, not a binding problem, which is exactly the

distinction the tightened anchor was added to make.

So per-element recall alone nearly re-saturates: strip out the counting beat and

it sits at 34/35. The discrimination in tier 5 comes mostly from joint

correctness (3 of 4 rather than 4 of 4) and from the composition anchor having

somewhere below 5 to land. Know that before quoting the pooled recall

figure as the headline.

The finding that matters more: the judge cannot count either

On the counting beat the judge reported nine acorns with the gold one fifth from

the left. It reported the same thing twice, in independent passes, once before a

counting discipline was added to the rubric and once after. A pixel count of that

image gives eight acorns, gold fourth from the left.

The judge landed the right pass or fail on every element of that beat, but only

by luck: the generator was also wrong, so "misbound" was the correct verdict for

the wrong reason. Had the model rendered exactly seven, the grader would most

likely have failed a correct render. That is a false negative on the lead model,

produced by the instrument rather than the subject.

The counting discipline added in response is still worth keeping, because of what

it did fix. It requires the judge to walk any countable group strictly left to

right, list every item with its color or size, and score counts and ordinals

against its own enumeration rather than against the checklist's claim. On the

handedness beat that discipline moved the verdict from 5/9 to 8/9 on the same

image: forced to enumerate, the judge correctly resolved that the raised hand on

the viewer's right side of the body is the character's own left hand, where

before it had contradicted itself and marked four elements misbound. Enumeration

repaired relational reasoning outright. It did not repair perception. Counting

from a glance is what vision models are worst at, and asking more carefully did

not change what the model saw.

t5-seven-acorns therefore carries a judgeCaveat in its definition, the runner

prints it before every render, and it is written into every result record. **Its

count and ordinal elements must not be quoted as judged data** until a

deterministic checker exists. The other four tier-5 beats carry no caveat; they

were judge-verifiable.

This generalises past this program. Any element the program scores by counting, on any

tier, inherits this error. The three-pots trap in tier 4 is a count of three,

inside the range where the judge is reliable, so experiment 5's numbers stand. Counts

above roughly five need a different instrument.

What this does not settle

  • n is 1. One replicate, one model, one arm. The probe shows the tier

discriminates; it does not measure anything.

  • One model. nano-banana has not seen tier 5. Experiment 5's degradation curve

predicts it should fall further and faster here, but that is a prediction.

  • Arm B only. No paragraph control, no reference arm.
  • Judge variance is visible. t5-size-order drew composition 4 on one pass

and 5 on the next, same image, the 4 attributed to an anatomy note that the

second pass did not raise. Roughly a point of noise on composition, which

argues for replicates before any composition delta under a point gets quoted.

  • The counting beat is on hold as described above.

Reproduction

# Full tier-5 battery, both live candidates, all arms.
node the beat-run script --tier=5 --model=gpt-image-2,nano-banana

# The validation probe, as run.
node the beat-run script --tier=5 --model=gpt-image-2 --arm=B \
  --replicates=1 --judge-via=claude

# Experiment 5's original slice is still the default and is unchanged.
node the beat-run script

Tier 5 is deliberately not in the default slice. An unqualified invocation still

reproduces experiment 5 rather than silently launching a larger, more expensive run.

Code

In the evaluation harness:

  • lib/beat-eval-set.mjs -- five tier-5 beats, the absence category, spelled

number words through twelve, BEAT_TIERS, beatKeysForTiers, and the

judgeCaveat field.

  • lib/beat-judge.mjs -- absence scoring semantics, the tier-5 counting

discipline and its enumeration output field, the tightened tier-5 composition

anchor, jointCorrect in recallStats, and renderBeatChecklist exported so

tests can assert every element reaches the judge.

  • the beat-run script -- --tier= selection defaulting to 1,4,

tier-derived run directory names, joint correctness in the console line and the

summary table, and the judge caveat printed and recorded.

  • tests/beat-eval-set.test.mjs -- 19 new tests, including one that pins every

tier-1 and tier-4 arm-B prompt against absence-driven drift.

Suite is green at 419 tests.

Next

  1. Run the full tier-5 battery on gpt-image-2 and nano-banana, arms A, B, and C,

two replicates, once the test-account token is refreshed so the judge goes back

through Bedrock. Roughly 60 renders.

  1. Build a deterministic counting checker, or accept human adjudication on count

elements, and unblock t5-seven-acorns.

  1. Take the cropping failure to the prompt-design side. It is the one tier-5

failure that looks like a prompting problem rather than a measurement problem.

Terminology, metrics, and methods

Every term the round pages use, in one place. Scores are 1-5 unless noted; all scored numbers on every page come from the same judge and are comparable to each other.

The judge

opus judge
Every score in the gallery comes from claude-opus-4-7, with the reference image (where one exists) and the rendered image(s) in the same request. It is deliberately a different model family than any generator under test. Scores from other judges are never mixed in -- a GPT-4o judge measures about a point more lenient on identical images, so cross-judge numbers are not comparable.
scored conservatively
The judge is instructed to score down when in doubt. Treat small differences between cells as noise; most cells are n=4, so directions matter, decimals do not.
reference gap (ref gap)
A flag, not a penalty: a character-sheet detail that the REFERENCE IMAGE itself never showed (e.g. feet cropped out of frame). The judge reports it separately and does not count its absence in a new render against continuity -- continuity is always judged against what the reference actually shows.

Reference machinery

frozen reference
The canonical image of a character, minted once and reused for every scene, every model, every round. Continuity is judged against this image. Frozen references are never regenerated -- if they changed between experiments, scores could not be compared.
style-matched reference
A frozen reference rendered IN the style being tested (e.g. maria-v3-graphic-novel.png for graphic-novel arms). Used on style experiments so the reference carries both identity and the style bar.
stable-image-core (the incumbent) -- NOT a candidate
Read every SIC row in this gallery as a measure of how far the current path falls short, not as a contender being scored. SIC is what production renders today and it is the best model we have measured at the target picker styles, graphic novel and 3D animation especially -- which is exactly why it mints our style exemplars. What it cannot do is render the story: experiment 5 measured 32% element recall, 45% of elements misbound, and 0 of 4 spatial elements in both arms. The program's goal is for gpt-image-2 and nano-banana to reach SIC's style quality while also getting the plot right and putting every named character on the page. SIC stays in the pipeline as the exemplar mint and comes out of it as the page renderer.
SIC exemplar / mint
A SIC exemplar is Stable Image Core's output for a character in one style -- on style experiments it serves as the frozen reference AND the visual definition of what the style should look like. This is the role SIC keeps.
character sheet
The fixed written list of a character's identity anchors (species/kind, colors, clothing, marks, proportions). The judge checks continuity anchor by anchor against it, subject to the reference-gap rule.
setting sheet
The same discipline applied to a room: a fixed description carrying five enumerable anchors (experiment 4's reading corner: round window with rain, red beanbag, oak bookshelf with brass globe, striped blue-and-yellow rug, chalkboard easel with a sun). 'The full five-anchor setting sheet verbatim' means this text rides in the prompt, unmodified, on every page.
mint-QC pass
Before a round runs, each sheet/beat is rendered once per model and vision-checked: does every anchor actually appear? Anchors that never render get rephrased. This exists because an early reference was minted missing a sheet anchor and the judge correctly dinged every scene for it.

Prompt arms

arm / cell
An arm is one prompting technique under test; a cell is one (model x arm) combination, usually rendered across 4 scenes or pages. Every experiment is control-vs-treatment: arm A is what production would naively do, later arms are candidate improvements, and the A-vs-B delta prices the improvement.
beat (story beat)
One moment in a story -- the smallest unit of "something happens." Borrowed from screenwriting. In this product each page is one beat and gets one picture: Maria handed Mr Penguin a yellow flower while he held a blue scarf. We say beat rather than paragraph because a paragraph is a chunk of text, while a beat is the content the picture is checked against -- who is in it, what they do, which objects, where.
ablation
Take a result apart to find which piece caused it. Borrowed from medicine, where it means removing tissue. Experiment 5 measured a 20-27 point win from "structured prompts," but that prompt changed four things at once (named each character's appearance, added labeled structure, spelled out counts, added a no-text clause), so the number could not say which one mattered. The ablation rebuilds the prompt one component at a time and watches where the gain first appears. Ours is the additive form -- start from nothing and add a rung per batch -- which is what the E5b ladder is. The answer: rung one did all of it.
structured-beat assembler
A production component that would rewrite a story beat into labeled parts (Actors / Action / Objects / Setting / Style) before sending it to the image model, instead of passing text close to the paragraph. Experiment 5 appeared to justify building one; the E5b ablation showed the entire win came from naming character appearance, which production already does, so it is not worth building for accuracy.
rung
One step of the ablation ladder -- the prompt with one more component than the step below it. Rungs are cumulative rather than factorial, so a ladder shows where a gain first appears but cannot detect an interaction between two components.
joint-correct
The share of renders where EVERY checklist element is right at the same time, as opposed to the percentage of elements right across all renders. It stays sensitive after per-element recall pins at 100%, which is why tier 5 reports it -- a model can get every noun right and still fail the beat.
rubric saturation
When the test stops discriminating because the model scores at the ceiling. gpt-image-2 hit exactly 100% recall and exactly 5.00 composition on every structured tier-1/4 cell, which sounds like success but means the rubric can no longer detect a regression on the model we intend to ship. The fix is harder test material (tier 5), not a different score.
idiom (per-model prompt idiom)
The prompting style a model family's own documentation says works best -- NOT generic prompt advice. For the gpt family: labeled Character/Scene/Style segments plus explicit exclusion clauses. For Gemini: narrative sentences leading with the style, semantic positives only, no exclusion lists. For Stable Image Core: caption-style text with weighted terms. Tuned and structured arms are always written in the target model's idiom; sending one family's idiom to another family is exactly the mistake the control arms measure.
sic-verbatim (control, style experiments)
The exact production Stable Image Core prompt -- weighted (term:1.4) syntax and all -- sent unmodified to the model under test, with only the scene clause appended. Measures what shipping today's prompt idiom unchanged would do.
tuned (treatment, style experiments)
A per-model prompt built in that model family's documented idiom: labeled Character/Scene/Style segments plus exclusion clauses for the gpt family; style-first narrative text with semantic positives for Gemini. Identity rides as a short lock (below) instead of the full sheet.
identity lock / 'the lock'
A one-clause identity pointer used in tuned arms -- 'the same girl as the attached reference image, with the orange glasses and grey jacket' -- instead of the full character sheet. Lets the reference image carry identity while the prompt spends its words on style and scene.
name-only (experiment 4 arm A)
The setting appears only as a name plus one clause ('in the cozy reading corner'). The control floor for setting continuity.
compact three-anchor lock (experiment 4 arm C)
The setting named with only its top three anchors. Tests whether a short lock buys the same room consistency as the full sheet -- experiment 4's answer was no (it holds its three anchors and drops the rest).
seed-hold (experiment 4 arm D, SIC only)
The full sheet plus the SAME random seed on all four pages (production derives a different seed per page). Only Stable Image Core exposes a seed. Buys trivially high consistency by making pages near-duplicates -- which also breaks recomposition and the night page, so it is not a real win.
paragraph-as-prompt (experiment 5 arm A)
The story paragraph sent as-is as the prompt. Approximates piping story text straight into the image model.
structured beat prompt (experiment 5 arm B)
The same story beat recomposed in the model's documented idiom, with emotion written as concrete facial/postural cues rather than emotion words.
reference-attention tax (experiment 5 arm C)
Arm B plus the characters' frozen references attached. Measures what reference conditioning costs in scene accuracy -- production will attach references, so this number connects experiment 5 back to the continuity experiments.

Metrics on the cards

arm
One prompting technique under test. Every experiment here is control-versus-treatment: arm A is what production would do naively, later arms are candidate improvements, and the gap between them is what the improvement is worth.
cell
One (model x arm) combination, rendered across the same four scenes. It is the unit every average is taken over -- and ours hold four images each, which is why direction matters more than decimals.
control arm
The arm standing in for current practice: what you would get without the change. Improvements are measured against it rather than in absolute terms.
anchors
The specific, checkable details that define a character or a room -- orange glasses, a red beanbag chair, an orange beak. The judge checks them one at a time rather than forming an overall impression, which is what makes a score auditable.
anchor-rich
A reference image that shows every one of a character's defining details clearly, nothing hidden or cropped. These are the references that produce reliable results; the opposite is what caused this program's one overturned conclusion.
dilution
What happens when a single sentence has to carry two characters' identities: neither gets described precisely enough, and whichever character has the weaker reference degrades first. The fix is one clause per character.
mint
To create a character or room reference for the first time. That image is then frozen and reused everywhere, so the quality of the mint caps everything measured downstream -- which is why reference QC is a production gate rather than a nicety.
exemplar
A reference image that doubles as the visual target: it defines both who the character is and what the style is supposed to look like. Stable Image Core mints ours.
matrix
The full grid of combinations an experiment covers -- characters x styles x models x arms x scenes. Multiplying it out is how the render count and cost are known before anything runs.
degenerate
Scores well on the metric while defeating the point of the test. The clearest case here: holding one random seed across four pages produces a perfect room-consistency score, but only because the pages come out as near-copies of each other.
continuity (character continuity)
Is this the same character the student built? 5 = unmistakably the same, every reference-visible anchor present; 3 = recognizably related but an anchor is missing or wrong; 1 = a different character. On pair cards the weaker character's match governs.
fidelity (scene fidelity)
Does the image show the requested setting, action, and elements? Judged against the scene prompt, not the reference.
style (style match)
How well the render realizes the REQUESTED style. On style experiments the reference deliberately shows the target style, and deliberate restyling is never counted against continuity.
cross-page consistency (cpc)
One score per 4-image set: does the character (or, experiment 4, the room) stay the same look render to render? 5 = identical on every page; 3 = noticeable drift on one page; 1 = changes between most pages.
anchor recall
Experiment 4's per-page checklist: each setting-sheet anchor marked present (✓), altered (△ -- there, but wrong shape/color/position), or absent (✗), with the judge's note. The percentage quoted in results docs is anchors-present across all pages of a cell.
same-room pairs
Experiment 4's product question, asked directly: for each adjacent page pair (1-2, 2-3, 3-4), would a reader say these two pictures happen in the same place? Reported as a count, e.g. 2/3.
night geometry / night palette
The night page keeps the room text identical and changes only the lighting clause, then gets two separate scores: did the room's LAYOUT survive the lighting change (geometry), and did the image actually go dark convincingly (palette)? Split so a pretty-but-rearranged night room cannot pass.
unprompted text
Did the model volunteer lettering nobody asked for (signs, book spines, gibberish captions)? Every sheet bakes in a blank-surface clause, so any lettering is the model's own contribution. Binary per page.
recall points
The unit deltas are quoted in. Element recall is a percentage, so a change in it is a difference in percentage points, not a percent change -- 74% to 100% is 26 recall points, not a 35% improvement. Two cautions when reading them: a point is worth more on a short checklist than a long one (one missed element out of four moves recall 25 points), and at n=8 cells per arm a few points is noise. Treat a 20-point gap as a signal and a 3-point gap as nothing.
element recall / binding (experiment 5)
Each story beat has a checklist of actors, actions, objects, counts, and spatial relations. The judge marks each element present, absent, or present-but-misbound (the umbrella exists but the wrong actor holds it). Element recall % is the headline; binding errors are counted separately because a misbound element is worse than a missing one.
cost pill ($ / image)
The practical price of one render as billed in that round: model output plus the reference image's input tokens where a reference is attached (measured on live billed calls, not sticker prices). Experiment 4 attaches no references, so its gpt-image-2 price is lower than the style experiments'.

Experiment map

Experiments 1-2 -- character continuity
Can a model take a frozen character reference and reproduce that character in new scenes? Experiment 2 added per-style exemplar references.
Experiments 3 / 3b -- style adherence
Control-vs-treatment prompt arms per model per style; 3b replicates experiment 3 on a second reference set. Combined finding: reference strength decides whether style-heavy prompting helps -- QC the reference first, tune prompts second.
Experiment 4 -- setting continuity
Same room across four recomposed pages, prompt-only, no references. Finding: the full setting sheet wins; short locks and name-only drift.
Experiment 5 -- scene accuracy
Element-by-element recall with binding checked, across a complexity ladder from simple beats to binding traps, plus the reference-attention tax. Finding: the failure is characters, not scenery -- actor recall 18-24% when the prompt is the raw story paragraph, 100% once the prompt states what the characters look like. gpt-image-2 reaches a clean 100% recall / 5.00 composition on every structured cell.
Experiment 6 -- multi-character
Experiment 3's protocol with two characters per image, judged in pair mode. Finding: pair continuity tracks the weaker member; no attribute swapping observed.

Story Studio style adherence -- experiment 3b

Findings
  • This held up when we ran it again. gpt-image-2 with tuned prompts was the best arm on every style in both the original and the repeat run, and tuning never lost ground when the reference image was a clean, single-subject shot of the character.
  • And this did not hold up, which changed the recommendation. The first run showed tuning hurting on 3D animation and collapsing nano-banana's vibrant-fantasy style. Neither reappeared once the reference images were cleaner, so both were artifacts of the first reference set rather than real effects. What decides whether style tuning helps is the quality of the reference image, not the style being requested.
  • Brad's 3D arm posted the largest tuning delta measured anywhere (3.50 → 5.00); fantasy-art is the hardest style (no 5.00 continuity in any arm).
  • New migration risk: baked-in mint scenery ((botanical garden:1.0)) fights the requested page scene and dings fidelity.
Recommended next steps
  • Ship pairing: gpt-image-2 + per-style tuned prompts + a reference-QC gate. nano-banana holds style as well and runs 3-4x faster, at the same all-in price.
  • Invest in reference QC (single subject, anchors visible) FIRST; per-style prompt tuning second -- at that point tuning is nearly free continuity.
  • Strip mint scenery from production prompts before appending page scenes.
Read the full written results for this experiment ↓method, every table, the reads, and the standing caveats

Experiment 3 repeated on a second reference set -- the replication that separated real effects from reference artifacts.

  • New references: redesigned Maria (short black hair, orange thick-rimmed glasses, grey jacket) plus Mr Penguin and Brad, minted with the heavier-weighted experiment 2 production prompts (weights up to (pixar:1.9) ride into the control arm verbatim).
  • Models: Maria ran all four; Mr Penguin and Brad run the lead pair only (gpt-image-2, nano-banana) per the demotion.
  • Same arms, same judge as experiment 3; the 1b seed variants are excluded by design.
  • Why it matters: whatever replicates across both reference sets is real; whatever doesn't was a reference artifact.
characters Maria (maria-v3), Mr Penguin (mr-penguin-v2)judge claude-opus-4-7 via Bedrockarms sic-verbatim vs tunedruns 2026-07-30/31
Graphic Novel

Graphic Novel

Maria

Maria Graphic Novel exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity5fidelity4style5cross page consistency4

All reference-visible anchors are preserved: Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, dark grey jacket open over an orange top, and blue jeans. Facial features match closely.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with short black hair, orange thick-rimmed glasses, dark grey open jacket over orange top, and blue jeans. Hair parting is slightly different (middle part vs reference's side-swept bangs), but all other anchors hold well.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with orange thick-rimmed glasses, orange top, dark grey open jacket, and blue jeans. Hair color is correct black, but appears longer/wilder than the short bob in the reference -- likely due to motion, but it is a minor anchor drift.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Asian girl with short black hair, orange thick-rimmed glasses, grey jacket over orange top, and blue jeans. Hair part is less distinctly 'down the middle' than reference but still central. All major anchors preserved.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency5
library
librarygpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency5

All reference-visible anchors preserved: Asian girl with short black bob and center-parted bangs, orange thick-rimmed glasses, dark grey/blue jacket open over orange top, blue jeans. Facial features and proportions match closely.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
beach
beachgpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency5

Same Asian girl with short black hair, orange thick-rimmed glasses, dark grey open jacket over orange top, and blue jeans -- all key anchors preserved. Minor drift: hair parting is less distinctly center-parted than the reference, but overall identity is unmistakable.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
sledding
sleddinggpt-image-2 · arm: tunedref gap
continuity5fidelity5style5cross page consistency5

All reference-visible anchors are preserved: short black hair with bangs/parted, orange thick-rimmed glasses, dark grey open jacket, orange top, blue jeans, and Asian features. Same character clearly recognizable.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
garden
gardengpt-image-2 · arm: tuned
continuity5fidelity5style4cross page consistency5

All reference-visible anchors hold: short black hair with bangs, orange thick-rimmed glasses, dark grey open jacket over orange top, and blue jeans. Facial features match closely.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity3style4cross page consistency4

Same character: short black hair with center-part bob, orange thick-rimmed glasses, dark grey jacket over orange top all present. Jeans visible. Minor drift: hair part is less pronounced/more of a straight fringe here versus the clearer center part in the reference, but overall identity is intact.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity5fidelity4style5cross page consistency4

All reference-visible anchors are preserved: Asian girl with short black hair and bangs, orange thick-rimmed glasses, dark grey jacket open over an orange top, and blue jeans. Facial proportions read as the same character.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with orange thick-rimmed glasses, dark grey jacket open over orange top, and blue jeans -- all key anchors present. Minor drift: hair appears slightly longer/less clearly parted down the middle than the reference's crisp bob with center part, but still recognizably Maria.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with short black hair parted down middle, orange thick-rimmed glasses, dark grey jacket open over orange top, and blue jeans. All key anchors preserved. Minor drift: hair looks slightly shorter/less framing the face, but identity holds.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same Asian girl with short black hair, orange thick-rimmed glasses, dark grey jacket open over orange top, and blue jeans -- all key anchors present. Hair parting is less distinctly center-parted than the reference's clear middle part, but overall identity is well preserved.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same Asian girl with short black bob, orange thick-rimmed glasses, dark grey jacket open over orange top, and blue jeans. Hair part is less distinctly centered than reference, but all major anchors hold.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity5fidelity5style4cross page consistency4

All reference-visible anchors are preserved: short black hair with center-parted bangs, orange thick-rimmed glasses, dark grey jacket worn open, orange top underneath, and blue jeans. Asian features consistent. Same character, just expressing joy.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity5fidelity4style4cross page consistency4

All reference-visible anchors are preserved: short black hair with center part, orange thick-rimmed glasses, dark grey jacket open over an orange top, and blue jeans. Facial features and proportions read as the same character.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

Mr Penguin

Maria Graphic Novel exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same penguin identity: black back, white belly, orange beak, orange webbed feet, wide curious eyes, and the yellow/orange cheek marking from the reference are all preserved. Minor drift: the beak appears slightly more open/expressive and the eye is rendered larger/more cartoonish, but the anchors hold.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The penguin is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same penguin species with black back, white belly, orange beak, orange webbed feet, and wide curious eyes all preserved. The yellow/orange marking near the eye is present in both. Minor drift: the black head stripe/mask pattern in the reference is simplified in the new image, and the eyes are rendered larger/more cartoonish, but identity clearly holds.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The penguin is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same penguin species with black back, white belly, orange beak, orange webbed feet, and wide curious eyes all preserved. The orange facial patch near the eye from the reference is less prominent/differently placed in the new image, and the proportions are chubbier/more cartoonish, but identity anchors largely hold.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The penguin is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same penguin species with black back, white belly, orange beak, orange webbed feet all preserved. Wide curious eyes are actually more clearly rendered here than in the reference. Minor drift: the yellow/orange cheek marking near the beak in the reference is less prominent, and the head shape is slightly rounder, but overall unmistakably the same character.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The penguin is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Same penguin identity: black back, white belly, orange beak, orange webbed feet, and the yellow/orange cheek patch from the reference are preserved. Eyes read as attentive though less 'wide curious' due to the downward reading gaze -- a minor drift.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
beach
beachgpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Same penguin identity: black back, white belly, orange beak, orange webbed feet all present. The distinctive orange patch near the eye/cheek and black head cap match the reference. Eyes read as curious but slightly less wide/round than reference. Overall unmistakably the same character.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
sledding
sleddinggpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Same penguin species with black back, white belly, orange beak, orange webbed feet, and the distinctive yellow/orange marking near the eye all preserved. Wide curious eyes are present though the open beak/excited expression differs slightly. Body proportions and markings match well.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
garden
gardengpt-image-2 · arm: tuned
continuity4fidelity4style5cross page consistency4

Same penguin species with black back, white belly, orange beak, orange webbed feet, and the distinctive yellow/orange marking near the eye all preserved. Wide curious eyes are present. Minor drift: the beak in the new image appears slightly more elongated/pointed, but overall identity is clearly maintained.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency3
library
librarynano-banana · arm: sic-verbatimref gap
continuity3fidelity5style3cross page consistency3

Same species (penguin) with black back, white belly, and orange feet preserved. However, the beak color drifted -- in the reference it is bright orange, while in the new illustration the beak is dark gray/black with only a hint of orange inside. The face marking also changed: the reference has a distinctive orange patch by the beak, absent in the new image. Proportions are also chubbier/more stylized-cute versus the reference's more realistic slender build.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The penguin is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency3

Same penguin species with black back, white belly, orange beak, orange webbed feet all preserved. The yellow/orange marking near the eye/cheek is retained. Wide curious eyes are more emphasized in the new image, which is fine. Minor drift: the new illustration adds a slight yellow tint on the belly that wasn't in the reference, but overall identity anchors hold.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The penguin is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity3fidelity4style3cross page consistency3

Both are penguins with black back, white belly, orange feet, and orange beak. However, the reference has a distinctive facial pattern with a yellow/orange marking near the eye and a more elongated adult body shape, while the new illustration shows a rounder, cartoonier penguin with a plain black head, no yellow eye marking, and different proportions. Recognizably related as a penguin but key facial anchor is lost.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The penguin is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency3

Same penguin species with black back, white belly, orange beak, orange webbed feet, and wide curious eyes. The yellow/orange marking near the eye is present in both. Minor drift: the reference shows a more prominent yellow patch beside the eye and a slimmer body, while the new image has a smaller yellow accent and a chunkier build, but identity holds.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The penguin is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency3
library
librarynano-banana · arm: tuned
continuity3fidelity5style4cross page consistency3

Same species with black back, white belly, orange beak, and orange webbed feet preserved. However, the reference penguin has a distinctive black-and-orange facial mask/marking around the eyes and a black chinstrap line, which is missing in the new illustration -- replaced with a plain black head and a yellow cheek patch. Eyes are wide and curious in both. Identity anchors partially drift.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity5style5cross page consistency3

Same penguin species with black back, white belly, orange beak, orange webbed feet, and wide curious eyes preserved. Minor drift: the new image adds a yellow/cream cheek patch near the beak that isn't present in the reference, where the face was more uniformly black with a small orange marking near the beak base.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tunedref gap
continuity3fidelity5style2cross page consistency3

Recognizably a penguin with black back, white belly, orange beak, and orange webbed feet. However, the reference clearly shows a distinctive facial pattern with a yellow/orange marking near the eye and a specific head shape reminiscent of a rockhopper or gentoo-like species, plus the wide curious eyes. The new illustration shows a generic cartoon penguin with closed/happy eyes, no eye marking, and a simpler rounder body -- several reference-visible anchors have drifted or been lost.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity3fidelity4style5cross page consistency3

Same species with black back, white belly, orange beak, and orange webbed feet preserved. However, a yellow/orange patch has appeared on the side of the head and on the belly/chest that was not present in the reference (which had a clean white belly and a small orange patch near the beak base only). The face markings have shifted, making it look like a slightly different penguin species.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

Brad

Maria Graphic Novel exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over white shirt, and black trousers -- all key reference-visible anchors survive. The black backpack is missing in the new illustration, a minor anchor drift.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The boy is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark hair, tan jacket over white shirt, and black trousers. Facial features match closely. The black backpack from the reference is missing in the new illustration, which is a minor anchor drift.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The boy is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket, white shirt, black trousers. Facial features match well. The black backpack visible in the reference is not shown in the new illustration, though this is understandable given the sledding action pose.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The boy is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark hair, tan jacket over white shirt, and black trousers. Facial features are consistent. Hair appears slightly wavier/more voluminous than the reference's shorter, neater cut, a minor drift. Backpack is not visible but he's kneeling so it could be off-frame/removed for the action.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The boy is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, same facial features, tan jacket over white shirt, black trousers, and black backpack straps visible. All reference-visible anchors preserved.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
beach
beachgpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Same Latino boy with tan jacket over white shirt, black trousers, and black backpack all present. Hair reads as dark and short but is noticeably wavier/more voluminous with a stylized swoop compared to the reference's shorter, flatter cut -- minor drift in one anchor.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
sledding
sleddinggpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

All reference-visible anchors preserved: same Latino boy with short dark wavy hair, tan jacket over white shirt, black trousers, and black backpack visible on shoulder. Facial features are consistent.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
garden
gardengpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: same Latino boy with short dark wavy hair, tan collared jacket over white shirt, black trousers, and black backpack. Facial features and proportions are consistent.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark hair, tan jacket over white shirt, and black trousers -- core anchors preserved. Hair appears slightly wavier/longer than reference's short cropped style, and the backpack is absent (though not needed in a reading scene). Face structure and skin tone match well.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The boy is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity3fidelity4style4cross page consistency4

Same tan jacket, white shirt underneath, black trousers, and dark wavy hair are preserved, and the Latino boy identity reads through. However, the face looks noticeably older/more mature than the reference's 9th-grade boy, and the hair is slightly longer/wavier. The backpack is present. Recognizably related but facial identity has drifted.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The boy is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket, white shirt underneath, black trousers, and black backpack straps visible. Hair is slightly wavier/longer than the reference's shorter cut, and the expression is joyful vs. neutral, but all key identity anchors are preserved.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The boy is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark hair, tan jacket over white shirt, black trousers, and black backpack straps visible. Hair reads slightly straighter/less wavy than reference, but overall identity anchors hold.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). (Graphic novel style:1.2), bold ink outlines, cel-shaded (comic-book:1.2). The boy is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity3fidelity4style4cross page consistency4

Same short dark wavy hair, similar face shape, tan jacket over white shirt, and black trousers are preserved. However, the black backpack visible in the reference is missing entirely in the new illustration, which is a notable identity anchor loss.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over white shirt, black trousers, and black backpack all present. Hair appears slightly wavier/more styled than the reference's shorter cut, a minor drift, but identity is clearly preserved.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity5fidelity5style4cross page consistency4

All reference-visible anchors held: same Latino boy with short dark wavy hair, tan jacket with same collar/zip/pocket detailing over white shirt, black trousers, and black backpack straps visible. Facial features and skin tone match.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity5fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over white shirt, black trousers, and black backpack straps visible. All reference-visible anchors preserved -- face shape, skin tone, and outfit match.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
3D Animation

3D Animation

Maria

Maria 3D Animation exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). Disney/Pixar 3D animation (pixar:1.9), warm color grading

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity5style5cross page consistency4

Same Asian girl with short black bob parted down the middle, orange thick-rimmed glasses, and orange top visible under jacket. Jacket color drifted from dark navy/black denim to grey (though sheet specifies grey, so this is actually closer to sheet). Blue jeans present in new image but not visible in reference. Identity is clearly preserved.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). Disney/Pixar 3D animation (pixar:1.9), warm color grading. The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with short black hair parted down the middle and distinctive orange thick-rimmed glasses. Orange top visible under the jacket, and blue jeans present. However, the jacket color drifted from dark navy/black in the reference to grey in the new illustration -- a minor but noticeable anchor drift.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). Disney/Pixar 3D animation (pixar:1.9), warm color grading. The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with black hair parted down the middle, orange thick-rimmed glasses, orange button-up shirt, grey/dark jacket worn open, and blue jeans -- all key anchors preserved. Hair appears slightly longer/more tousled in the new image versus the tidy bob in the reference, a minor drift.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). Disney/Pixar 3D animation (pixar:1.9), warm color grading. The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Asian girl with short black hair parted down middle, orange thick-rimmed glasses, and orange top all preserved. Jacket color drifted from dark navy/black in reference to grey in new image, which is a minor anchor drift but matches the sheet's 'dark grey jacket' description.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). Disney/Pixar 3D animation (pixar:1.9), warm color grading. The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency5
library
librarygpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency5

All reference-visible anchors preserved: Asian girl with short black bob parted down the middle, orange thick-rimmed glasses, dark grey open jacket over orange button-up top, and blue jeans visible. Facial features match closely.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tunedref gap
continuity5fidelity5style5cross page consistency5

All reference-visible anchors hold: short black hair with center part and bob shape, orange thick-rimmed glasses, dark grey open jacket over orange top, and now visible blue jeans. Face structure and skin tone consistent.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency5

All reference-visible anchors are preserved: Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, dark grey jacket open over an orange button-up top, and blue jeans are visible. Facial structure is consistent.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned
continuity5fidelity4style5cross page consistency5

All reference-visible anchors hold: short black hair with center part and bob length, orange thick-rimmed glasses, dark grey open jacket over orange collared top. Same Asian girl with matching facial features. Jeans visible where shown.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity5fidelity4style5cross page consistency4

All reference-visible anchors are preserved: Asian girl with short black bob parted down the middle, orange thick-rimmed glasses, dark grey open jacket over orange top, and jeans. Facial structure and proportions read as the same character.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). Disney/Pixar 3D animation (pixar:1.9), warm color grading. The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with short black bob parted down the middle, orange thick-rimmed glasses, dark grey open jacket over orange top, and blue jeans -- all key anchors preserved. Minor drift: hair looks slightly less voluminous and the face is a bit more stylized/older-looking, but identity clearly holds.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). Disney/Pixar 3D animation (pixar:1.9), warm color grading. The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with orange thick-rimmed glasses, orange button-up top, dark jacket worn open, and blue jeans -- all key anchors preserved. Hair appears slightly longer/less bob-shaped than the reference's clean chin-length bob, a minor drift, but overall unmistakably the same character.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). Disney/Pixar 3D animation (pixar:1.9), warm color grading. The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Asian girl with short black hair parted down the middle, orange thick-rimmed glasses preserved. Jacket color drifted from dark navy/black in the reference to grey in the new illustration, and the orange top under the jacket is no longer clearly visible. Face shape and overall identity read as the same character.

promptIn a (botanical garden:1.0) a 5th grader Asian girl wearing a jeans and a grey jacket, wearing (orange thick-rimmed glasses:1.5), short black hair parted down middle (black hair:1.7). Disney/Pixar 3D animation (pixar:1.9), warm color grading. The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: Asian girl with short black bob parted down the middle, orange thick-rimmed glasses, dark grey/navy open jacket over orange collared top, similar facial proportions. Identity is unmistakably the same character.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tunedref gap
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: short black hair with middle part and bob length, orange thick-rimmed glasses, dark grey open jacket over orange top, Asian features, similar face shape and proportions. Blue jeans now visible (unseen in reference but consistent with sheet).

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tunedref gap
continuity4fidelity5style5cross page consistency4

Same Asian girl with black hair, orange thick-rimmed glasses, dark jacket open over orange top, and now blue jeans (not visible in reference but matches sheet). Hair length appears slightly longer/less bobbed than reference, and the jacket reads more as dark denim/navy than the dark grey shown in reference, a minor drift.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tunedref gap
continuity4fidelity5style5cross page consistency4

Same Asian girl with short black hair parted in middle, orange thick-rimmed glasses, dark grey jacket open over orange top all preserved. Face shape and features match well. Blue jeans now visible (not seen in reference but consistent with sheet). Minor drift in hair styling but identity is clearly maintained.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

Mr Penguin

Maria 3D Animation exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptA proportional penguin with wide,curious eyes in a (botanical garden:1.0). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity5fidelity5style5cross page consistency4

All reference-visible anchors preserved: black back, white belly, orange beak, orange webbed feet, wide curious eyes with the same warm brown iris, small tuft of hair on head, and same rounded proportions. Unmistakably the same penguin.

promptA proportional penguin with wide,curious eyes in a (botanical garden:1.0). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading. The penguin is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity5fidelity4style5cross page consistency4

All reference-visible anchors present: black back, white belly, orange beak, orange webbed feet, wide curious eyes, and the small tuft of hair on top of the head is preserved. Same character clearly.

promptA proportional penguin with wide,curious eyes in a (botanical garden:1.0). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading. The penguin is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity5fidelity5style5cross page consistency4

All reference-visible anchors hold: black back, white belly, orange beak, orange webbed feet, wide curious eyes, and the same little tuft of head feathers. Same character unmistakably.

promptA proportional penguin with wide,curious eyes in a (botanical garden:1.0). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading. The penguin is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity5fidelity5style5cross page consistency4

All reference-visible anchors preserved: black back, white belly, orange beak, orange webbed feet, wide curious eyes, small tuft of hair on top, same chubby proportions. Unmistakably the same character.

promptA proportional penguin with wide,curious eyes in a (botanical garden:1.0). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading. The penguin is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

All reference-visible anchors preserved: black back, white belly, orange beak, orange webbed feet, wide curious dark eyes, and the same little tuft of feathers on the head. Facial proportions and expression match closely.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

Same penguin: black back, white belly, orange beak, orange webbed feet, wide curious brown eyes, and the signature tuft of hair on top of the head all carry over faithfully.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

Same penguin: black back, white belly, orange beak, orange webbed feet, wide curious dark eyes with the same eye shape, and the distinctive little tuft of hair on top of the head. All reference-visible anchors preserved.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: black back, white belly, orange beak, orange webbed feet, wide curious dark eyes, and the small tuft of feathers on top of the head. Body proportions and cute chubby build match closely.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same penguin species with black back, white belly, orange beak, orange webbed feet, and wide curious eyes. The signature tuft of hair on top of the head is preserved. Minor drift: the beak appears slightly shorter/more rounded and the overall proportions are a bit more compact, but identity clearly holds.

promptA proportional penguin with wide,curious eyes in a (botanical garden:1.0). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading. The penguin is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity5fidelity3style5cross page consistency4

Same penguin: black back/head, white belly, orange beak, wide curious brown eyes, small tuft of hair on head, and orange feet visible. All reference-visible anchors preserved.

promptA proportional penguin with wide,curious eyes in a (botanical garden:1.0). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading. The penguin is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity5fidelity4style5cross page consistency4

Same penguin: black back, white belly, orange beak, orange webbed feet, wide curious eyes, and the distinctive little tuft of head feathers all carry over faithfully.

promptA proportional penguin with wide,curious eyes in a (botanical garden:1.0). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading. The penguin is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same penguin character: black back, white belly, orange beak, orange webbed feet, wide curious eyes, and the signature tuft of hair on top all preserved. Minor drift: the black head area appears slightly more brownish/grey in the new image versus solid black in reference, and the facial mask shape is a touch different.

promptA proportional penguin with wide,curious eyes in a (botanical garden:1.0). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading. The penguin is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency5
library
librarynano-banana · arm: tuned
continuity5fidelity5style5cross page consistency5

The penguin retains all reference-visible anchors: black back, white belly, orange beak, orange webbed feet, wide curious dark eyes, small tuft of hair on head, and the same rounded chubby body proportions. Identity is unmistakable.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity5fidelity5style5cross page consistency5

Same penguin: black back/head, white belly with same facial mask shape, orange beak, orange webbed feet, wide curious dark eyes, and the same little tuft of hair on top of the head. All reference-visible anchors preserved.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity5fidelity5style5cross page consistency5

All reference-visible anchors are preserved: black back and head, white belly, orange beak, orange webbed feet, wide curious dark eyes, small tuft of hair on top of head, and the same rounded chubby proportions. Unmistakably the same penguin.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity5fidelity4style5cross page consistency5

Same penguin: black back, white belly, orange beak, orange webbed feet, wide curious brown eyes, and the little tuft of hair on top of the head all match the reference. Proportions and appealing chubby shape preserved.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

Brad

Maria 3D Animation exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (trousers:1.2). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading.

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, thick eyebrows, brown eyes, tan jacket over a shirt with a blue graphic, and dark trousers. Facial structure and expression match closely. The black backpack is not visible, but that's understandable for a seated indoor reading scene.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (trousers:1.2). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading.. The boy is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity3fidelity4style5cross page consistency4

Same face, hair, eyebrows, tan jacket, and dark trousers all carry over well. However, the black backpack visible in the reference is missing in the new illustration, which is a notable identity anchor drop.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (trousers:1.2). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading.. The boy is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity3fidelity4style5cross page consistency4

Same face, hair, eyebrows, tan jacket, and black trousers carry over well. However, the backpack visible in the reference is missing in the new illustration, which is a reference-visible anchor that broke.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (trousers:1.2). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading.. The boy is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, thick eyebrows, brown eyes, tan jacket over a shirt with a small blue graphic, and dark trousers. Face and outfit match well. The black backpack visible in the reference is not shown in the new illustration, a minor anchor drift.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (trousers:1.2). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading.. The boy is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

Same boy: short dark wavy hair, thick brows, brown eyes, same facial structure and smile. Tan jacket over graphic tee with same blue emblem, black trousers, black backpack straps visible. All reference-visible anchors preserved.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

Same boy: short dark wavy hair, thick eyebrows, brown eyes, same facial structure and smile. Tan jacket over shirt with same emblem, black trousers, and dark backpack all present.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, thick eyebrows, brown eyes, tan jacket over graphic tee, black trousers, and black backpack all preserved from the reference.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, thick eyebrows, brown eyes, tan jacket over graphic shirt, black trousers, and black backpack straps visible. All reference-visible anchors preserved.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, thick eyebrows, tan jacket over a shirt, dark trousers, and backpack straps visible. Face shape and expression match well. Minor drift: the shirt graphic is less clearly visible and jacket appears slightly more orange-toned, but overall identity anchors hold.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (trousers:1.2). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading.. The boy is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity4fidelity3style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket, black trousers, and black backpack all present. Face shape and features are consistent. Minor drift: the shirt underneath is less visible/different, and he appears barefoot (context-appropriate for beach), but core identity anchors hold.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (trousers:1.2). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading.. The boy is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over a shirt, black trousers, and backpack straps visible. Facial features (brows, eyes, nose) match well. Minor drift: the shirt graphic appears slightly different but still bluish, and the jacket looks a touch lighter/more uniform, but overall identity holds.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (trousers:1.2). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading.. The boy is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Latino boy: short dark wavy hair, tan jacket, black trousers, black backpack, similar shirt with graphic all present. Facial features look consistent (brown eyes, thick brows). Minor drift in facial proportions -- eyes appear slightly larger/rounder in the new illustration, but identity clearly holds.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (trousers:1.2). Disney/Pixar 3D animation (pixar:1.5), vivid warm color grading.. The boy is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, thick eyebrows, brown eyes, tan jacket over a shirt, black trousers, and a black backpack (visible beside chair). Facial identity and outfit hold well. Minor drift: the jacket in the new image reads more like a button-up shirt-jacket than the structured zip-up vest-jacket in the reference, but overall identity anchors are preserved.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, thick eyebrows, brown eyes, tan jacket over shirt, black trousers, and black backpack. Identity anchors hold well. Minor drift: the shirt underneath appears teal/greenish rather than the grey with blue graphic seen in reference, and the jacket looks slightly more like a full jacket than the vest-over-jacket layered look.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, thick brows, tan jacket over blue-graphic shirt, black trousers, and backpack straps visible. Identity anchors hold well. Minor drift: the shirt graphic and jacket details are slightly simplified, and the face reads a touch younger, but recognizably the same character.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket, dark trousers, and black/dark backpack. Face shape and features are consistent. Minor drift: trousers appear more olive/khaki than the black in the reference, and the shirt underneath is less clearly visible, but overall identity holds.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
Vibrant Fantasy

Vibrant Fantasy

Maria

Maria Vibrant Fantasy exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity4style5cross page consistency4

Same Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, and orange top visible under jacket. Jeans present. The jacket color drifted from dark black/navy in reference to a lighter grey, which is a minor anchor drift but actually aligns with the sheet.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity5style4cross page consistency4

Same Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, and orange top under a jacket with blue jeans. However, the jacket color drifted from black/dark navy in the reference to light grey in the new illustration -- a minor anchor drift, though it matches the sheet's 'dark grey' description.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity4style4cross page consistency4

Same Asian girl with black hair, orange thick-rimmed glasses, orange top, and jeans preserved. Jacket is grey rather than the near-black of the reference, but that matches the sheet. Hair looks slightly longer/messier due to motion but still short bob with bangs. The reference's small star hair clip is not visible in the new image, a minor anchor drift.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity5style5cross page consistency4

Same Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, orange top, and blue jeans all preserved. The jacket shifted from black in the reference to dark grey in the new illustration -- one minor color anchor drifted, though the sheet actually specified grey so it's a reference-vs-sheet discrepancy. Star hair clip from reference is absent.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: short black hair with center-parted bob and star hair clip, orange thick-rimmed glasses, dark jacket open over orange top, blue jeans. Face structure and Asian features match.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tunedref gap
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: short black hair with side-parted bangs and star hair clip, orange thick-rimmed glasses, dark jacket open over an orange top, and blue jeans. Facial features and Asian ethnicity are consistent.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Same Asian girl with orange thick-rimmed glasses, orange top, dark jacket open, blue jeans, and the star hair clip is preserved. Hair appears slightly longer/wavier than the reference's clean short bob, a minor drift, but identity clearly holds.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

All reference-visible anchors hold: short black hair with bangs and star hair clip, orange thick-rimmed glasses, dark jacket open over orange top, blue jeans. Same Asian girl clearly recognizable.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, orange top, and dark jacket. The jacket reads more as a grey hoodie than the reference's black bomber, and jeans shifted from bright orange-accented to plain dark blue -- minor drift but identity clearly holds.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatimref gap
continuity3fidelity4style2cross page consistency4

Same Asian girl with short black hair parted down middle and orange thick-rimmed glasses is preserved. However, the jacket color changed from dark navy/black in the reference to grey in the new illustration, and the star hair clip visible in the reference is missing. The orange top and jeans are retained.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity4style4cross page consistency4

Same Asian girl with short black hair parted in the middle and orange thick-rimmed glasses is clearly present. Orange top peeks out under the jacket and blue jeans are retained. However, the jacket is a puffy grey winter coat rather than the dark grey open bomber-style jacket from the reference -- a minor clothing drift appropriate to the winter scene but still a change.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity3fidelity3style4cross page consistency4

Same Asian girl with short black hair parted with bangs and orange thick-rimmed glasses is preserved. However, the outfit changed notably: the reference shows a dark jacket over an orange top, while the new image shows a grey hoodie over a green top. Two clothing anchors (jacket color shifted to grey and inner top changed from orange to green) drifted, so identity is recognizable but multiple anchors broke.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same character clearly recognizable: short black hair with bangs, orange thick-rimmed glasses, orange top under dark jacket, blue jeans, and even the star hair clip is preserved. Jacket reads more black than dark grey but that's minor and consistent with reference.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same Asian girl with short black hair, orange thick-rimmed glasses, dark grey jacket open over orange top, and blue jeans. Star hair clip from reference is preserved. Hair is slightly shorter/less full but still recognizably the same character.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same Asian girl with black hair, orange thick-rimmed glasses, dark jacket open over orange top, and blue jeans - all key anchors preserved. The star hair clip from the reference is also present. Hair appears slightly longer/less crisply bobbed than reference, a minor drift.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tunedref gap
continuity4fidelity4style4cross page consistency4

Same Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, dark jacket open over orange top. The star hair clip from the reference is preserved. Minor drift: the jacket in the reference reads as very dark navy/black rather than dark grey, and the new image maintains that dark tone consistently. Jeans not visible due to framing but not contradicted.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

Mr Penguin

Maria Vibrant Fantasy exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same penguin identity: black/blue back, white belly, orange beak, orange webbed feet with claws, wide curious eyes, and similar iridescent jewel-tone shading on the head. Minor drift: the belly lost the gradient rainbow sheen seen in the reference and reads as plainer white, but all core anchors hold.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The penguin is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity4style4cross page consistency4

Same penguin identity: black/blue iridescent back, white belly, orange beak, orange webbed feet, wide curious eyes, and the distinctive white facial curve. The belly lost the green/blue gradient seen in reference, but that's a minor drift attributable to the scene/lighting change.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The penguin is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity4fidelity5style4cross page consistency4

Same penguin identity: black/blue iridescent back, white belly, orange beak, orange webbed feet, wide curious eyes all preserved. The jewel-tone purple/blue sheen on the plumage carries over. Minor drift: expression is now open-mouthed and excited rather than calm profile, but this is pose/emotion, not identity.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The penguin is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same penguin identity: black/blue back, white belly, orange beak, orange webbed feet, and wide curious eyes are all present. The jewel-tone purple/blue iridescence on the back matches the reference. Minor drift: the eyes are more front-facing/cartoonish and larger than the reference's more subtle eye, but this is a stylistic tweak rather than an identity break.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The penguin is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

All reference-visible anchors preserved: black/blue back, white belly, orange beak with same shape, orange webbed feet, wide curious eyes, and the distinctive white cheek/eye stripe from the reference all appear intact.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Same penguin identity: black back, white belly, orange beak, orange webbed feet all preserved. The wide curious eye is slightly less prominent/rounder in the new image (eye reads a bit smaller and more focused), but species, colors, and markings including the white cheek curve match well.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: black/dark blue back, white belly with the same gradient hint of green/yellow, orange beak, orange webbed feet with black claws, and the same wide curious eye with white cheek marking. Head shape and coloration match unmistakably.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Same penguin identity preserved: black/blue back, white belly, orange beak, orange webbed feet, and the distinctive white curve around the eye all match. Eyes read as curious though slightly less wide than reference. Minor drift in beak shape (more pointed) but overall unmistakably the same character.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity3fidelity5style4cross page consistency4

Same species (penguin) with orange beak, orange feet, white belly, and wide curious eyes preserved. However, the back/head coloring drifted -- the reference shows a predominantly black/dark blue back with a distinct white facial marking, while the new illustration shows a more purple/blue head without the same white cheek stripe. The overall build is chubbier and more stylized.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The penguin is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same penguin identity: black/blue iridescent back, white belly, orange beak, orange webbed feet, wide curious eyes all present. Color palette leans slightly more purple/violet than the reference's blue-black, but the jewel-tone treatment and all key anchors carry over.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The penguin is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity3fidelity5style4cross page consistency4

Same penguin species with black back, orange beak, orange feet, and blue iridescent head sheen carried over. However, the belly gained a distinct yellow/green gradient patch that wasn't as prominent in the reference (which had more of a subtle rainbow shimmer on a mostly white belly). The wide curious eyes are preserved but the expression is more open-mouthed/excited. Recognizably the same character but belly coloration drifted.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The penguin is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: same penguin with iridescent jewel-toned black/blue back, white belly with green-cyan gradient, orange beak, orange webbed feet, and wide curious eyes. The facial markings and body proportions match closely.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The penguin is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Same penguin identity preserved: black/blue back, white belly, orange beak, orange webbed feet with claws, wide curious eyes, and the same iridescent blue-purple sheen on the head. Minor drift: the reference shows a distinct white curved marking around the eye/cheek area which is less pronounced here, and the beak shape is slightly stubbier, but overall clearly the same character.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same penguin identity preserved: black/dark blue back with iridescent sheen, white belly, orange beak, orange webbed feet, and the distinctive white cheek marking from the reference. The wide curious eye is present. Minor drift: the belly lacks the multi-color gradient of the reference and the plumage looks slightly softer, but core anchors hold.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity4fidelity5style3cross page consistency4

Same penguin species with black back, white belly, orange beak, orange webbed feet, and wide eyes all preserved. The iridescent blue/purple sheen on the black feathers carries over from the reference. Minor drift: a yellow cheek patch appears in the new illustration that wasn't clearly in the reference, and the body proportions are chubbier/more stylized.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity4fidelity4style5cross page consistency4

Same penguin identity clearly preserved: black/dark back with iridescent blue-purple sheen, white belly, orange beak, orange webbed feet, and the distinctive white curve on the face. Wide curious eyes present though slightly less prominent. Minor drift in that the back has more purple/magenta emphasis, but this is consistent with the reference's own jewel-tone sheen.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

Brad

Maria Vibrant Fantasy exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing (tan jacket:1.32) and black trousers (black trousers:1.3). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatimref gap
continuity5fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, same facial features, same tan jacket over black shirt, same black trousers. All reference-visible anchors preserved.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing (tan jacket:1.32) and black trousers (black trousers:1.3). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The boy is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity4style3cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, and black trousers -- all reference-visible anchors preserved. Facial features are consistent. Minor drift: the jacket in the new image looks more satin/leather-like versus the matte cotton in the reference, but identity is unmistakable.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing (tan jacket:1.32) and black trousers (black trousers:1.3). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The boy is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, black trousers -- all reference-visible anchors preserved. Facial features are consistent. Minor drift: the black backpack from the sheet isn't visible, but it wasn't in the reference either, so not penalized.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing (tan jacket:1.32) and black trousers (black trousers:1.3). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The boy is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatimref gap
continuity5fidelity5style5cross page consistency4

Same Latino boy with identical short dark wavy hair, facial features, tan jacket over black shirt, and black trousers. All reference-visible anchors preserved.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing (tan jacket:1.32) and black trousers (black trousers:1.3). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The boy is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tunedref gap
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, and black trousers. Facial structure and hairstyle match well. The backpack was not visible in the reference either, so no penalty there. Minor drift in jacket detailing but overall identity holds.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tunedref gap
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark hair, tan jacket over black shirt, and black trousers. Hair reads slightly more curly/wavy than reference but still consistent. Facial structure and skin tone match well. Backpack was not visible in reference but appears here - not a penalty. One minor drift: hair texture appears curlier.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tunedref gap
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, and black trousers all preserved. The backpack from the sheet is not visible in the reference, so its faint presence/absence isn't a continuity issue. Facial features read as the same person; minor drift in jacket detailing due to style.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tunedref gap
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, and black trousers all present. The backpack was not visible in the reference, so its addition here is neutral. Facial features and hairstyle match well; minor drift in jacket detailing but identity holds.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatimref gap
continuity4fidelity3style4cross page consistency4

Same Latino boy with short dark hair, tan jacket over black shirt, and black trousers -- all reference-visible anchors preserved. Hair appears slightly less voluminous/wavy and the expression is softer with a smile, but identity clearly holds. Backpack from sheet not visible but also absent in reference.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing (tan jacket:1.32) and black trousers (black trousers:1.3). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The boy is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatimref gap
continuity4fidelity4style4cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, and black trousers. Facial features and hairstyle match well. The backpack from the sheet is not visible, but it wasn't in the reference either, so continuity holds on all reference-visible anchors.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing (tan jacket:1.32) and black trousers (black trousers:1.3). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The boy is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatimref gap
continuity4fidelity4style4cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, black trousers -- all reference-visible anchors held. Minor drift: facial proportions look slightly younger/more animated, but identity is intact. Backpack from sheet is not visible in reference either.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing (tan jacket:1.32) and black trousers (black trousers:1.3). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The boy is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatimref gap
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, and black trousers -- all key anchors from the reference are preserved. The black backpack from the sheet isn't visible, but it wasn't in the reference either. Minor drift: hair looks slightly shorter/tighter on the sides, but identity clearly holds.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing (tan jacket:1.32) and black trousers (black trousers:1.3). Vibrant fantasy art, jewel-tone palette (3D Vibrant fantasy:1.65). The boy is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tunedref gap
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: same Latino boy with short dark wavy hair, same tan jacket over black shirt, black trousers. The black backpack, absent in reference, appears here as an added element. Facial features and identity read as unmistakably the same character.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tunedref gap
continuity4fidelity5style4cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, black trousers. The black backpack, absent from the reference, is now added per the sheet. Facial identity and clothing anchors visible in the reference are preserved well; minor drift is that he's barefoot and in a kneeling pose, but identity holds.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tunedref gap
continuity4fidelity5style4cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, black trousers. Backpack is now added (absent from reference but listed in sheet). Facial identity is reasonably preserved though slightly younger/more animated. Minor drift in hair styling but core anchors hold.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tunedref gap
continuity4fidelity5style5cross page consistency4

Same Latino boy with short dark wavy hair, tan jacket over black shirt, and black trousers all preserved. Facial features are close. The black backpack is added here (not visible in reference) but matches the sheet. Minor drift in jacket detail but identity clearly holds.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
Fantasy Art

Fantasy Art

Maria

Maria Fantasy Art exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity5fidelity4style5cross page consistency4

Same Asian girl with short black hair parted down the middle, orange thick-rimmed glasses, dark grey jacket open over an orange top, and blue jeans. All reference-visible anchors are preserved.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity4style4cross page consistency4

Same Asian girl with short black hair, orange thick-rimmed glasses, grey jacket open over orange top, and blue jeans -- all reference anchors present. Hair parting is slightly less centered and hair looks a touch longer/wavier, a minor drift.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity3fidelity5style4cross page consistency4

Same Asian girl with orange thick-rimmed glasses and orange top under a dark jacket, plus blue jeans -- core anchors survive. However, hair is noticeably longer and looser (not the short bob parted down the middle from the reference), and the jacket reads as a puffy quilted coat rather than the structured denim-style jacket. Recognizably related but a couple of anchors drifted.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style4cross page consistency4

Same Asian girl with short black hair parted down middle, orange thick-rimmed glasses, dark grey jacket open over orange top, and blue jeans. All key anchors preserved. Minor drift: hair appears slightly longer/less bob-shaped than reference, but identity is clearly maintained.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Same Asian girl with short black bob with bangs, orange thick-rimmed glasses, orange top, dark jacket open, blue jeans. The jacket reads more as dark denim/navy rather than dark grey, a minor drift, but overall identity anchors are intact.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Same girl: short black hair with bangs, orange thick-rimmed glasses, dark jacket open over orange top, blue jeans all preserved. Hair appears slightly longer/side-parted rather than middle-parted, a minor drift.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned
continuity3fidelity5style5cross page consistency4

Same Asian girl with orange thick-rimmed glasses, orange top, dark jacket open, and blue jeans -- key anchors held. However, the hair is notably longer and flowing rather than the short bob seen in the reference, which is a clear drift on a visible anchor.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency4

All reference-visible anchors hold: Asian girl with short black bob and bangs, orange thick-rimmed glasses, dark grey/navy open jacket, orange top underneath, blue jeans. Facial features closely match.

promptCharacter: the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity4style4cross page consistency4

Same Asian girl with short black bob parted down the middle, orange thick-rimmed glasses, dark grey jacket over orange top, and blue jeans. All key anchors preserved. Minor drift: the jacket appears less distinctly denim-textured and the bangs are slightly less pronounced, but identity is intact.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity4fidelity5style4cross page consistency4

Same Asian girl with short black hair parted down middle, orange thick-rimmed glasses, dark grey jacket open over an orange top, and blue jeans. All key anchors preserved. Minor drift: hair looks slightly longer/less structured than reference bob, but identity is clearly maintained.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Asian girl with short black hair, orange thick-rimmed glasses, and jeans. The jacket shifted from a dark denim/black open jacket over an orange t-shirt to a grey puffer over an orange/rust knit sweater -- the orange top and grey jacket concept are preserved but the specific garments drifted slightly.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity5style4cross page consistency4

Same Asian girl with short black bob parted down middle, orange thick-rimmed glasses, dark jacket over orange top, and blue jeans. All key anchors preserved. Minor drift: the jacket in the reference has visible orange/copper buttons and a more textured denim-like appearance, while the new one is smoother, but identity is unmistakable.

promptIn a (botanical garden:0.7) a 5th grader Asian girl wearing a (jeans:1.1) and a grey jacket, wearing (orange thick-rimmed glasses:1.57), short black hair parted down middle (black hair:1.7). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity5fidelity5style4cross page consistency4

All reference-visible anchors are preserved: Asian girl with short black bob and center-parted bangs, orange thick-rimmed glasses, dark grey/navy jacket open over an orange top, and blue jeans. Facial proportions and youthful appearance match.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity4style5cross page consistency4

Same Asian girl with short black bob, orange thick-rimmed glasses, dark grey jacket open over orange top, and blue jeans -- all key anchors present. Hair parting is less clearly centered and slightly longer/messier than reference, a minor drift.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same Asian girl with black hair, orange thick-rimmed glasses, dark grey/navy jacket open over orange top, and blue jeans. Hair appears slightly longer/wavier than the reference's short bob with straight bangs, a minor drift, but all other anchors hold.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same Asian girl with short black bobbed hair with bangs, orange thick-rimmed glasses, dark grey/denim jacket open over orange top, and blue jeans. All key anchors preserved. Minor drift: hair parting is less clearly center-parted and jacket looks slightly more denim-textured, but identity is unmistakable.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same girl as the attached reference image -- short black hair, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

Mr Penguin

Maria Fantasy Art exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same penguin identity: black back, white belly, orange beak with black tip, black-and-white facial markings match the reference. Eyes appear curious though slightly less wide. Feet are visible with orange/claw tones consistent with reference. Minor drift in eye expression but overall recognizably the same character.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The penguin is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity5style4cross page consistency4

Same penguin species with black back, white belly, orange beak, and the distinctive white facial stripe/band pattern all preserved. Wide curious eyes present. Minor drift: the feet appear more grayish/muted rather than the vivid orange seen in the reference, though a hint of orange remains.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The penguin is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same penguin species with black back, white belly, orange beak, and wide curious eyes preserved. The distinctive white eye-stripe/facial marking from the reference is less pronounced in the new image (more of a small white brow accent), and the feet appear more orange-brown/clawed similarly. Overall identity holds with minor drift in facial markings.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The penguin is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style4cross page consistency4

Same penguin: black back, white belly, orange beak, orange webbed feet, and the distinctive white eye-stripe/facial markings all carry over. Wide curious eyes present. Minor drift: the reference shows a more slender build with dirty/muddy flipper tips, while the new image shows a plumper body with cleaner flippers, but identity anchors hold.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The penguin is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency5
library
librarygpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency5

Same penguin: black back, white belly, orange beak, orange webbed feet with visible claws, wide curious eyes, and the same facial markings (white eye stripe pattern) all carry over faithfully.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tuned
continuity4fidelity5style4cross page consistency5

Same penguin identity: black back, white belly, orange beak, wide curious eye with white facial markings all preserved. Feet are orange/yellow as in reference. One minor drift: the distinctive black chin-strap line visible on the reference's neck is less pronounced, and the belly lacks the speckling of the reference, but overall unmistakably the same character.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency5

Same penguin: black back, white belly, orange beak, orange webbed feet, and the distinctive white facial stripe/eye markings from the reference are preserved. Proportions and wide curious eyes carry over.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned
continuity5fidelity4style5cross page consistency5

Same penguin: black back, white belly, orange beak with dark tip, distinctive white eye stripe/facial pattern, orange feet with dark claws, and wide curious eyes all preserved from the reference.

promptCharacter: the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency3
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity5style4cross page consistency3

Same penguin species with black back, white belly, orange beak, and the distinctive white facial marking around the eyes is preserved. Wide curious eyes present. Minor drift: feet are less vividly orange and partially hidden, and the beak is slightly shorter/stubbier than reference, but overall unmistakably the same character.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The penguin is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity4fidelity4style4cross page consistency3

Same penguin species with black back, white belly, orange beak, and the distinctive white facial marking curving around the eye is preserved. Wide curious eyes present. Orange webbed feet visible but less saturated/prominent than reference. Overall unmistakably the same character with only minor drift in feet color intensity.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The penguin is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity3fidelity5style3cross page consistency3

Same species (penguin) with black back, white belly, orange beak, and wide eyes preserved. However, the reference shows a distinctive facial marking (white curved stripe around the face/eye area typical of a Magellanic-style penguin) that is absent in the new illustration -- the new penguin has a simpler face with just a small white cheek patch. Also, the new penguin looks chubbier/more cartoonish in proportion versus the slender adult proportions of the reference. Orange feet preserved.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The penguin is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity4style4cross page consistency3

Same penguin species with black back, white belly, orange beak, orange webbed feet, and the distinctive white facial band around the black head are all preserved. Wide curious eyes are present and even more emphasized. Minor drift: the dark claws/talons remain, and proportions are slightly more cartoonish/anthropomorphic (holding objects with wing-hands), but identity anchors hold.

promptA penguin with wide,curious eyes and clear face in a (botanical garden:1.0). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The penguin is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Same penguin species with black back, white belly, orange beak, white facial markings around eyes, and curious expression preserved. Feet are orange with visible claws matching reference. Minor drift: belly has a slight cream/yellow tint and the feet appear a bit more clawed/stylized, but overall unmistakably the same character.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same penguin identity: black back, white belly, orange beak, orange webbed feet, and the distinctive white facial marking around the eye all carry over. Eyes read as curious though smaller/less wide than reference. Minor drift: a yellow tint appears on the upper chest that wasn't in the reference.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same penguin species with black back, white belly, orange beak, orange webbed feet, and wide curious eyes all present. The facial markings (white eye stripe) carry over. Minor drift: the new illustration adds a yellowish tint to the chest and has a chubbier, more cartoony proportion, but identity anchors hold.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Same penguin identity: black back, white belly, orange beak, orange webbed feet with visible claws, white eye markings all preserved. The wide curious eye is slightly less pronounced/rounder brown eye vs the reference's bolder eye, a minor drift, but overall unmistakably the same character.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same penguin as the attached reference image -- black back, white belly, orange beak, orange webbed feet, wide curious eyes, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

Brad

Maria Fantasy Art exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity4style5cross page consistency4

Same Latino boy with short dark wavy hair, same tan bomber jacket with patch on sleeve and embroidered design, teal shirt underneath, pendant necklace, and black trousers all match the reference. The black backpack from the sheet is not visible in the reference either, so not penalized. Minor drift: facial features are very close but slightly softer.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The boy is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity5style4cross page consistency4

Same Latino boy with short dark wavy hair, tan bomber jacket with shoulder patch and emblem, teal shirt, pendant necklace, and black trousers all carry over. Hair reads slightly curlier in the new image, but identity anchors hold. Backpack from sheet is not visible in reference so not penalized.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The boy is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity4style4cross page consistency4

Same Latino boy with short dark wavy hair, tan bomber jacket with patch on sleeve and emblem on chest, teal shirt, pendant necklace, and black trousers all reproduced faithfully. The black backpack from the sheet isn't visible, but it also isn't clearly visible in the reference, so this is a reference gap rather than continuity break.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The boy is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatimref gap
continuity5fidelity4style4cross page consistency4

Same Latino boy with short dark wavy hair, same tan bomber jacket with badge/patch on left sleeve and emblem on right chest, teal/green shirt underneath, pendant necklace, black trousers. All reference-visible anchors preserved.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The boy is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency5
library
librarygpt-image-2 · arm: tunedref gap
continuity4fidelity4style5cross page consistency5

Same Latino boy with short dark wavy hair, tan bomber jacket with patch on sleeve, teal shirt underneath, pendant necklace, and black trousers all preserved. The backpack was not visible in the reference but appears added here. Facial features and proportions match well.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tunedref gap
continuity4fidelity5style5cross page consistency5

Same Latino boy with short dark wavy hair, tan bomber jacket with shoulder patch, teal shirt underneath, pendant necklace, and black trousers all reproduced faithfully. The backpack (sheet anchor) is now present, which wasn't in the reference. Facial features and proportions match well.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tunedref gap
continuity4fidelity5style4cross page consistency5

Same Latino boy with short dark wavy hair, tan bomber jacket with patch/emblem detail, teal shirt underneath, necklace, and black trousers all present. Backpack is now visible (was not in reference but is on the sheet). Facial features are consistent. Minor drift in jacket detailing but identity holds.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tunedref gap
continuity4fidelity5style5cross page consistency5

Same Latino boy with short dark wavy hair, tan bomber jacket with dark collar and sleeve pocket patch, teal/dark green shirt underneath, and black trousers. Facial features match well. The backpack is now visible (was not clearly shown in reference), which is consistent with the sheet. Minor drift: the emblem/patch on the jacket chest looks slightly different in design, but overall identity holds.

promptCharacter: the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a semi-realistic digitally painted fantasy artwork with finely detailed material textures, dramatic dynamic lighting, lifelike anatomical proportions, and vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. No flat cartoon, no chibi or exaggerated proportions, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatimref gap
continuity4fidelity4style4cross page consistency4

Same Latino boy with short dark wavy hair, tan bomber jacket with black collar/ribbing, teal shirt underneath, pendant necklace, and black trousers all carry over. Facial features match well. The backpack from the sheet isn't visible, but it wasn't in the reference either. Minor drift in jacket patch details.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The boy is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatimref gap
continuity4fidelity4style4cross page consistency4

Same Latino boy with short dark hair, tan bomber jacket with patch, teal shirt, necklace, and black trousers all preserved. Hair reads slightly less wavy/more groomed but the identity clearly holds. Backpack from the sheet is not visible in the reference either, so not counted.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The boy is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatimref gap
continuity4fidelity4style4cross page consistency4

Same Latino boy with short dark wavy hair, tan bomber jacket with patch detail, dark teal shirt, pendant necklace, and black trousers all carry over. Backpack from the sheet isn't visible, but it wasn't in the reference either. Minor drift: face looks slightly younger/more animated, but identity holds.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The boy is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatimref gap
continuity4fidelity4style4cross page consistency4

Same Latino boy with short dark wavy hair, tan bomber jacket with sleeve patch, teal shirt underneath, necklace, and black trousers all preserved. Backpack from the sheet is not visible in reference either, so not penalized. Minor drift: the shirt graphic/embroidery on jacket is slightly different, but overall identity holds.

promptIn a (botanical garden:0.5) one single 9th grade Latino boy wearing tan jacket and black trousers (black trousers:1.2). Depicted in (Fantasy Art:1.5) with detailed textures, dynamic lighting, and lifelike proportions. Vivid, colorful, cinematic quality.. The boy is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same Latino boy with short dark wavy hair, tan bomber jacket with patch detail, teal shirt underneath, pendant necklace, and black trousers all preserved. The black backpack appears beside him rather than worn, which is a minor drift but still present.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity5style4cross page consistency4

Same Latino boy with short dark wavy hair, tan bomber jacket with dark collar and arm patch, black trousers all preserved. The teal shirt underneath is barely visible but hinted at. The pendant necklace from the reference is not visible, a minor anchor drift.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tunedref gap
continuity4fidelity5style4cross page consistency4

Same Latino boy with short dark wavy hair, tan bomber jacket with patch detail, teal shirt, pendant necklace, and black trousers all carry over. The backpack is added (not visible in reference but on sheet). Face reads as the same character, though slightly younger/rounder due to the joyful expression.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tunedref gap
continuity4fidelity5style4cross page consistency4

Same Latino boy with short dark wavy hair, tan bomber jacket with sleeve pocket patch, dark shirt underneath, necklace, and black trousers all preserved. Backpack is added (not visible in reference but listed in sheet). Facial features are close but slightly younger/softer, a minor drift.

promptAn illustration with a semi-realistic digitally painted fantasy artwork: finely detailed textures, dramatic dynamic lighting, lifelike proportions, vivid cinematic color -- Fantasy art with detailed textures, dynamic lighting, and lifelike proportions. It shows the same boy as the attached reference image -- short dark wavy hair, tan jacket, black trousers, black backpack, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.

Full written results

the complete document behind this page -- method, every table, the reads, and the standing caveats

Last updated: 2026-07-31

Experiment 3b -- style adherence with the second reference set: running results

Experiment 3 repeated with the "Reference Photos Experiment 2" SIC set (the 1.ref-maria_* images; the 1b seed variants deliberately excluded). Same design as experiment 3 -- three styles x four models x two arms (sic-verbatim control vs per-model tuned treatment), maria's style-matched exemplar as the frozen reference per arm, opus judge throughout, gallery updated after each style.

What changed vs experiment 3:

  • New character design (maria-v3): short black hair parted down the middle, orange thick-rimmed glasses, dark grey jacket open over an orange top, blue jeans. The glasses are the sharpest small-mark anchor of the whole evaluation -- exactly the drift class (Pepper's star, Zuri's frames) that separates arms. The backpack shows only as straps in two exemplars (soft anchor); the vibrant-fantasy exemplar's star hair clip is style-specific and excluded from the sheet.
  • Much heavier production weight syntax in the control arm: the experiment 2 SIC prompts carry weights up to (pixar:1.9), (black hair:1.7), (orange thick-rimmed glasses:1.5) -- shipped verbatim, which sharpens the control arm's question: does heavy weighted syntax confuse models that do not document weight parsing?
  • Seeded mints: the exemplars were generated at fixed seeds (logged in the SIC prompt file), so the reference set itself is reproducible.
  • Exemplar quality is higher than experiment 3's: single subject, portrait-framed, consistent identity across all three styles -- SIC held this design noticeably better than the experiment 3 one.

Status: 24 runs dispatched 2026-07-30 with inline opus judging (credentials live this time -- no deferred re-judge stage). Sections fill in per style.

Results: graphic-novel (judged 2026-07-30, opus, inline)

ModelA: sic-verbatimB: tunedCont deltaA styleB style
gpt-image-14.254.250.005.004.75
gpt-image-1-mini4.004.000.004.754.75
gpt-image-24.254.75+0.505.004.75
nano-banana4.254.50+0.254.754.00

Experiment 3's direction replicates, at smaller magnitude. Tuning still wins on the strong pair (gpt-image-2 +0.50, nano +0.25; experiment 3 saw +0.75/+1.00) and is flat on the weaker pair. gpt-image-2 tuned is again the best arm (4.75, consistency 5).

Why the deltas shrank: the control arm got better, not the treatment worse. This round's SIC prompts carry the identity anchors WITH heavy weights -- (orange thick-rimmed glasses:1.5), (black hair:1.7) -- so the control prompt is itself doing identity work the experiment 3 control (which named almost nothing about the character) did not. Control continuity rose to 4.00-4.25 across all four models. The weighted syntax still caused no visible artifacts or text leakage on any model.

New control-arm artifact, worth carrying into the migration read: several control cells took fidelity hits (3s on library/beach scenes) because the SIC prompt hard-codes (botanical garden:1.0) and "jewel-tone themed room" while the scene clause asks for a library or beach -- the two prompt halves fight, and at weight 1.0 the garden sometimes wins. That is exactly what would happen if production concatenated its current character prompt with new scene text: the exemplar-mint scenery leaks into pages. The tuned arms, which carry no baked-in scenery, took no such hits (fidelity 4.75+ everywhere).

Results: 3d-animation (judged 2026-07-30, opus, inline)

ModelA: sic-verbatimB: tunedCont deltaB consistency
gpt-image-14.004.000.004
gpt-image-1-mini4.004.000.004
gpt-image-24.005.00+1.005
nano-banana4.254.50+0.254

Style adherence: 5.00 in essentially every arm, both variants.

This overturns experiment 3's 3d-animation finding. Experiment 3 measured tuned prompts HURTING the strong pair on this style (gpt-image-2 -0.75, nano -0.50), and the working hypothesis was that rendering-property language fights the reference near the model's native look. With this experiment's cleaner reference set, the same tuned prompts gave gpt-image-2 +1.00 -- a second perfect arm (5.00 continuity, 5.00 style, consistency 5) -- and nano +0.25. The experiment 3 negative deltas were apparently conditional on that round's weaker 3d exemplar (anchors hidden, identity thin), not a property of the style or the prompt language. Downgrade the experiment 3 hypothesis accordingly: with a reference that carries identity strongly, property-rich style language does not compete with it. This non-replication is the single most valuable output of running the round twice -- the per-style prompt-table recommendation now needs reference quality as a third axis, and the honest statement is that prompt-tuning effects are contingent on reference strength, measured at n=4 both times.

Control-arm scene conflict appeared again (fidelity dips where (botanical garden:1.0) fights the requested scene), same class as graphic-novel's.

Results: vibrant-fantasy (judged 2026-07-30, opus, inline)

ModelA: sic-verbatim (cont/style)B: tuned (cont/style)Cont deltaStyle delta
gpt-image-13.50 / 3.503.75 / 4.75+0.25+1.25
gpt-image-1-mini3.75 / 4.003.75 / 5.000.00+1.00
gpt-image-24.00 / 4.504.75 / 5.00+0.75+0.50
nano-banana3.50 / 3.754.00 / 4.00+0.50+0.25

Tuned wins or ties every cell on both dimensions -- and experiment 3's nano style collapse did not replicate (experiment 3: tuned nano style fell to 3.00; here it EDGED UP, 4.00 vs 3.75). gpt-image-2 tuned is once again the best arm (4.75 continuity, 5.00 style, 5.00 fidelity). The control arm's style adherence sagged for the gpt family (3.50-4.50) -- the heavy (3D Vibrant fantasy:1.65) term in this SIC prompt lands as an instruction to go 3D, and the gpt models obeyed it at the expense of the painterly brief, which the tuned prompts' explicit painterly language corrected (+1.25/+1.00 style).

Mr Penguin extension (judged 2026-07-31, opus; lead pair only per the model demotion)

The 1.ref penguin set, same protocol. His per-style exemplars disagree with each other about his design far more than Maria's (chibi with a head tuft in 3d, realistic with yellow nape patches in graphic-novel, iridescent in vibrant-fantasy), so he tests the reference-strength axis from the wobbly end: each arm's reference is internally strong, but the character has no stable cross-style identity beyond the minimal sheet.

StyleModelA: sic-verbatim (cont/style)B: tuned (cont/style)Cont delta
graphic-novelgpt-image-24.00 / 5.004.00 / 5.000.00
graphic-novelnano-banana3.50 / 4.00 (cons 3)3.25 / 4.00 (cons 3)-0.25
3d-animationgpt-image-25.00 / 5.005.00 / 5.000.00
3d-animationnano-banana4.50 / 5.005.00 / 5.00 (cons 5)+0.50
vibrant-fantasygpt-image-24.00 / 4.504.50 / 5.00+0.50
vibrant-fantasynano-banana3.75 / 4.504.00 / 4.25+0.25

Reads: 3d-animation is his easy style -- the chibi 3d exemplar is a simple, strong identity, and gpt-image-2 held it perfectly in BOTH arms (a control-arm perfect score, the first anywhere in experiments 3/3b -- with a reference this clean, even the raw production prompt suffices). Graphic-novel is his hard style, and it is nano that pays: 3.25-3.50 continuity with consistency 3 and a slightly negative tuned delta -- the one exemplar whose design departs most from the others, echoing the reference-strength story rather than contradicting the tuning story. Vibrant-fantasy lands in between, with tuned modestly positive on both models and gpt-image-2 again the best arm (4.50/5.00). Deltas overall are much flatter for the penguin than for Maria: when the reference is either very simple (3d) or design-ambiguous (gn), prompt tuning has little left to add or subtract. Penguin totals across styles (A -> B): gpt-image-2 4.33 -> 4.50, nano-banana 3.92 -> 4.08 -- direction matches Maria's round, magnitude halved.

Brad extension (judged 2026-07-31, opus; lead pair only)

The 1.ref Brad set: a ninth-grade Latino boy in a tan jacket and black trousers, mid-strength references (the mint prompts run the botanical garden at only 0.5, and the vibrant-fantasy exemplar hides several sheet anchors -- the judge flagged reference_gap on every vibrant-fantasy arm). His control arms were re-rendered after the SIC-prompt correction (the log had stored his 2.ref evolved wordings under the experiment 1 ids), so both arms below ran the true experiment 1 prompts.

StyleModelA: sic-verbatim (cont/style)B: tuned (cont/style)Cont delta
graphic-novelgpt-image-24.00 / 5.004.75 / 5.00+0.75
graphic-novelnano-banana3.75 / 4.754.25 / 4.50+0.50
3d-animationgpt-image-23.50 / 5.005.00 / 5.00+1.50
3d-animationnano-banana4.00 / 5.004.00 / 5.000.00
vibrant-fantasygpt-image-24.50 / 4.504.00 / 5.00-0.50
vibrant-fantasynano-banana4.00 / 4.254.25 / 4.50+0.25

Cross-page consistency is 4 on every arm.

Reads: the +1.50 on gpt-image-2 3d-animation is the largest single tuning delta of the entire evaluation -- and it lands exactly where the reference-strength model predicts the middle of the range should be: a real human character on a decent-but-not-pristine reference, where the control prompt leaves the most on the table. The one negative delta (gpt-image-2 vibrant-fantasy, -0.50) is also where the reference is weakest -- every vibrant-fantasy arm carries reference_gap flags, so the judge is scoring off a thin anchor set and aggressive style language has room to tip identity, same mechanism as experiment 3. Brad totals across styles (A -> B): gpt-image-2 4.00 -> 4.58, nano-banana 3.92 -> 4.17 -- the same direction as Maria and Mr Penguin, with gpt-image-2 again taking the bigger tuning payoff.

Fantasy-art extension (judged 2026-07-31, opus; all three characters, lead pair only)

The fourth style, added after the main matrices: Fantasy art with detailed textures, dynamic lighting, and lifelike proportions -- the style block whose brief pushes hardest against cartoon simplification, and the first style where "lifelike proportions" is part of the ask.

CharacterModelA: sic-verbatim (cont/style)B: tuned (cont/style)Cont delta
Mariagpt-image-24.00 / 4.254.00 / 5.000.00
Marianano-banana4.00 / 4.254.25 / 4.25+0.25
Mr Penguingpt-image-24.00 / 4.504.75 / 4.75 (cons 5)+0.75
Mr Penguinnano-banana3.75 / 3.75 (cons 3)4.00 / 4.50+0.25
Bradgpt-image-24.25 / 4.254.00 / 4.75 (cons 5)-0.25
Bradnano-banana4.00 / 4.004.00 / 4.000.00

Reads: this is the hardest style yet for both models -- no arm reached 5.00 continuity and control style adherence sagged to 3.75-4.50, the widest gap below the bar since experiment 2's fantasy-illustration laggard. Tuning wins or ties every cell on style (Maria's tuned gpt-image-2 arm is the only 5.00 anywhere in the section) and stays flat-to-positive on continuity, with one small negative on Brad -- whose fantasy-art reference carries the same reference_gap flags as his vibrant-fantasy set, so the thin-reference caveat applies again. The weakest arm in the section is nano's control on Mr Penguin (3.75/3.75, consistency 3), continuing his graphic-novel pattern: when the exemplar's design departs from the character's other references, nano pays first. gpt-image-2 tuned is the best arm on all three characters, now across four styles.

Across experiments 3 and 3b: two reference sets, two Maria designs, 192 judged renders on the Maria matrices plus the Mr Penguin and Brad lead-pair extensions and the 48-render fantasy-art extension, same arms, same judge.

Findings that replicated (treat as real):

  1. gpt-image-2 with tuned prompts is the best arm on every style in both experiments -- three perfect or near-perfect arms in 3b alone (5.00 gn... 4.75 gn, 5.00 3d, 4.75 vf) on top of experiment 3's perfect graphic-novel. The ship pairing is not in question.
  2. Tuned prompts never lose on the strong models when the reference is strong. Every negative tuned delta in the whole experiment traces to experiment 3's weaker reference set.
  3. The raw production prompt (weighted syntax and all) never breaks -- no parse artifacts, no text leakage, serviceable continuity everywhere -- but it consistently leaves continuity and style points on the table, and it carries a real scene-conflict risk: baked-in mint scenery ((botanical garden:1.0), "jewel-tone themed room") fights the requested page scene and periodically wins, dinging fidelity. Production prompt assembly must strip mint-scenery before appending page scenes.
  4. Zero moderation blocks in 100+ "Disney/Pixar" renders across four models and both experiments. Still not a guarantee; empirically the block rate remains 0.

Experiment 3 findings that FAILED to replicate (downgrade them):

  1. The "property language fights the reference near the model's native look" effect (experiment 3's negative 3d deltas) and nano's vibrant-fantasy style collapse both vanished with the cleaner reference set. They were reference-quality artifacts, not properties of styles or prompt idioms.
  2. The corrected model: reference strength is the moderating variable. With a single-subject, anchor-rich reference (experiment 3b's set), aggressive style language helps or is neutral everywhere; with a weaker reference (experiment 3's set), it can tip identity. The production stack should therefore invest in reference QC FIRST (single subject, portrait framing, anchors visible -- the mint-QC loop from the prompting research), and per-style prompt tuning second -- at which point tuning is close to a free win.

Standing caveats: n=4 per cell in both experiments; one character per round; opus style scores and Mason's eye disagreed once before (experiment 2), so the tuned renders still deserve a human pass against the exemplars before any of this ships.

Cost of experiment 3b: ~$3.20 in renders plus judge tokens; cumulative experiment 3 + 3b: ~$6.50.

Story Studio style adherence -- experiment 3

Findings
  • The complaint that opened this ticket is answered. Graphic novel had been rendering as flat cartoon rather than comic-book art. Describing the style by its properties -- ink line weight, hard-edged cel shadows, flat saturated fills -- instead of just naming it produced genuine comic interior art, and gpt-image-2's tuned arm scored 5 out of 5 on continuity, style, and cross-page consistency: the first perfect cell in the program.
  • Tuning looked style-dependent: big wins on inked styles (+0.75/+1.00), negative on 3D for the strong models, and nano's long tuned prompt collapsed vibrant-fantasy style to 3.00.
  • The raw production prompt never broke -- but left up to a full continuity point on the table.
Recommended next steps
  • Replicate on a second, cleaner reference set before trusting the negative 3D deltas (this became experiment 3b -- and they did NOT replicate; they were reference artifacts).
  • Keep nano's style language short and dense; long narrative dilutes its style signal.
Read the full written results for this experiment ↓method, every table, the reads, and the standing caveats

The experiment that answers "graphic novel looks cheap" -- judge scores said styles matched, the eye said flat cartoon.

  • Setup: each style's Stable Image Core exemplar is both the identity reference and the visual bar.
  • Models: gpt-image-2 and nano-banana. Two older gpt tiers were also run in this experiment and have since been retired -- gpt-image-1 came to cost more per image than the model that outscores it, and the mini tier ignored prompt quality entirely.
  • Arm A (control): the exact production SIC prompt, sent unmodified -- weighted syntax and all.
  • Arm B (treatment): a per-model prompt written in that family's documented idiom.
  • The measurement: the per-model A-vs-B delta = what prompt tuning is worth.
character Maria (maria-v2)judge claude-opus-4-7 via Bedrockarms sic-verbatim vs tunedruns 2026-07-30
Graphic Novel

Graphic Novel

Maria Graphic Novel exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint prompt(SIC prompt not found)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue jacket open over purple top, similar face shape and small earrings. The maroon-pink backpack visible in the reference is absent in the new illustration, a minor anchor drift, but otherwise unmistakably the same character.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Graphic novel style, bold ink outlines, cel-shaded (analytical-curious:0.6). The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue open jacket, and purple top -- all reference-visible anchors present. The maroon-pink backpack is missing in the new image, a minor anchor drop.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Graphic novel style, bold ink outlines, cel-shaded (analytical-curious:0.6). The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue open jacket, and purple top -- all key reference-visible anchors held. The maroon-pink backpack is not visible in the sledding image, a minor anchor drift, but otherwise unmistakably the same character.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Graphic novel style, bold ink outlines, cel-shaded (analytical-curious:0.6). The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: long straight black hair with blue-black highlights, same facial features, teal-blue jacket worn open over purple top, and the maroon-pink/red backpack straps are visible. Same character unmistakably.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Graphic novel style, bold ink outlines, cel-shaded (analytical-curious:0.6). The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency5
library
librarygpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency5

All reference-visible anchors preserved: long straight black hair, teal-blue open jacket over purple top, maroon-pink backpack strap visible, same Asian facial features and skin tone. Character is unmistakably the same girl.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
beach
beachgpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency5

All reference-visible anchors present: long straight black hair with same blue-black highlights, teal-blue jacket worn open, purple top underneath, maroon-pink backpack straps visible, same Asian facial features and proportions.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
sledding
sleddinggpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency5

All reference-visible anchors preserved: long straight black hair, teal-blue jacket worn open, purple top underneath, maroon-pink backpack straps visible, same Asian features and skin tone. Unmistakably the same character.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.
garden
gardengpt-image-2 · arm: tuned
continuity5fidelity5style5cross page consistency5

All reference-visible anchors are preserved: long straight black hair with the same blue-black highlights, same facial features (eyes, nose, lips), teal-blue jacket open over purple top with matching button details, and the maroon-pink backpack. Same age and proportions.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: bold ink outlines with varied line weight, hard-edged cel shading with deep shadow shapes, flat saturated color fills, printed comic book interior art finish -- Graphic novel style, bold ink outlines, cel-shaded. No flat minimal cartoon, no clipart look, no soft airbrushed shading, no speech bubbles, panels, or lettering.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue open jacket over purple top, and maroon-pink backpack straps visible. All key anchors present. Minor drift in facial proportions but unmistakably the same character.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Graphic novel style, bold ink outlines, cel-shaded (analytical-curious:0.6). The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue open jacket, purple top, and maroon-pink backpack straps visible. Face shape and features match well. Minor drift in hair length appearing slightly longer, but all key anchors preserved.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Graphic novel style, bold ink outlines, cel-shaded (analytical-curious:0.6). The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity4style4cross page consistency4

Same Asian girl with long straight black hair, teal-blue open jacket, purple top, and maroon-pink backpack -- all reference-visible anchors preserved. Facial proportions look slightly younger/more cartoony but identity holds.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Graphic novel style, bold ink outlines, cel-shaded (analytical-curious:0.6). The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue jacket open over a purple top -- all key reference-visible anchors present. The maroon-pink backpack is not visible in the new illustration, a minor anchor drift, but overall identity is clearly preserved.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Graphic novel style, bold ink outlines, cel-shaded (analytical-curious:0.6). The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency5
library
librarynano-banana · arm: tuned
continuity5fidelity5style4cross page consistency5

All reference-visible anchors are preserved: long straight black hair with the same dark blue-black tone, same Asian facial features, teal-blue jacket worn open over a purple top, and the maroon-pink backpack is present beside her. Identity is unmistakably the same character.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity5fidelity5style5cross page consistency5

All reference-visible anchors are preserved: long straight black hair, teal-blue open jacket, purple top underneath, maroon-pink/red backpack, and Asian features consistent with the reference.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tuned
continuity5fidelity5style4cross page consistency5

All reference-visible anchors preserved: long straight black hair, Asian features, teal-blue jacket worn open, purple top underneath, and maroon-pink backpack straps visible. Same character clearly recognizable.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tuned
continuity5fidelity4style5cross page consistency5

Same long straight black hair, same Asian girl facial features, teal-blue jacket worn open, purple top underneath, and maroon-pink backpack all present and consistent with the reference.

promptAn illustration with the confident inked linework and dramatic hard-edged cel shadows of printed comic book interior art, every surface a flat saturated fill bounded by crisp black ink -- Graphic novel style, bold ink outlines, cel-shaded. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
3D Animation

3D Animation

Maria 3D Animation exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint prompt(SIC prompt not found)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with shoulder-length dark hair, teal-blue open jacket over a purple top with floral detail -- matches reference closely. The maroon-pink backpack is not visible in the new illustration, but she's seated so it could be behind her; hair length is similar (medium, not clearly long/straight as sheet says but matches reference). Minor drift only.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Disney/Pixar 3D animation, warm color grading (analytical-curious:0.6). The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with shoulder-length dark hair, teal-blue jacket open over a purple top, maroon-pink backpack visible, khaki pants preserved. Hair appears slightly shorter/wavier than reference but overall identity anchors hold.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Disney/Pixar 3D animation, warm color grading (analytical-curious:0.6). The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity4style5cross page consistency4

Same Asian girl with dark hair, teal-blue open jacket, purple top with floral pattern, and matching circular badge visible. Hair appears slightly longer/wavier in the new image versus the shorter bob in the reference -- minor drift. Backpack is not visible but plausibly hidden by pose/winter gear.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Disney/Pixar 3D animation, warm color grading (analytical-curious:0.6). The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatim
continuity4fidelity5style5cross page consistency4

Same Asian girl with dark hair, teal-blue open jacket over purple top preserved. Face shape and expression match well. The maroon-pink backpack visible in reference is not shown in the new illustration (could be occluded by angle), and hair appears slightly darker/more black vs. the reference's dark brown, but overall identity is clearly the same character.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Disney/Pixar 3D animation, warm color grading (analytical-curious:0.6). The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: tunedref gap
continuity3fidelity5style5cross page consistency4

Same Asian girl with teal-blue jacket, purple floral top, and pink backpack -- outfit anchors hold well. However, hair length changed significantly: the reference shows shoulder-length wavy/bobbed hair, while the new illustration shows long straight hair past the shoulders. This is a visible identity anchor drift from the reference.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tunedref gap
continuity3fidelity5style5cross page consistency4

Same Asian girl with teal-blue open jacket over purple top and maroon-pink backpack -- key clothing anchors preserved. However, the reference clearly shows medium-length wavy/bob hair, while the new illustration gives her long straight black hair. This is a visible anchor drift from the reference.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tuned
continuity4fidelity5style5cross page consistency4

Same Asian girl with dark hair, teal-blue open jacket, purple top, and maroon-pink backpack -- all key anchors present. Minor drift: reference shows shorter/bobbed hair while the new image shows longer flowing hair, though this could partly be motion.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tuned
continuity3fidelity5style5cross page consistency4

Same Asian girl with teal-blue open jacket over purple floral top and pink/maroon backpack -- these anchors hold. However, hair length changed significantly: the reference shows shoulder-length wavy/bobbed dark brown hair, while the new image shows long straight black hair. This is a notable identity drift on a key anchor visible in the reference.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a glossy 3D animated feature film still with soft global illumination, warm cinematic color grading, subsurface-scattered skin, and gentle depth of field -- Disney/Pixar 3D animation, warm color grading. No flat 2D rendering, no outline strokes, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatimref gap
continuity4fidelity5style4cross page consistency4

Same Asian girl with dark hair, teal-blue jacket worn open, purple top visible underneath, and pink backpack strap visible. Hair appears slightly shorter/similar bob length as reference (reference showed medium-length hair, not truly long). Facial features and overall identity are consistent.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Disney/Pixar 3D animation, warm color grading (analytical-curious:0.6). The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatimref gap
continuity4fidelity3style5cross page consistency4

Same Asian girl with dark shoulder-length hair, teal-blue open jacket, purple top underneath, and pink backpack straps visible. The circular badge on the jacket is preserved. Hair appears slightly shorter/less voluminous than reference but overall identity holds.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Disney/Pixar 3D animation, warm color grading (analytical-curious:0.6). The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatim
continuity4fidelity4style5cross page consistency4

Same Asian girl with dark hair, teal-blue jacket open over purple floral top with badge, and pink backpack straps visible. Hair appears slightly shorter/less straight than reference (which showed shoulder-length wavy hair, matching well). Facial identity is consistent. Minor drift in hair length/styling but core anchors hold.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Disney/Pixar 3D animation, warm color grading (analytical-curious:0.6). The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatimref gap
continuity4fidelity5style5cross page consistency4

Same Asian girl with shoulder-length dark hair, teal-blue jacket worn open over purple top with floral detail and circular badge, pink/maroon backpack strap visible. Hair appears slightly shorter than reference but overall identity anchors hold well.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Disney/Pixar 3D animation, warm color grading (analytical-curious:0.6). The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity3fidelity5style4cross page consistency4

Recognizably an Asian girl with dark hair, teal jacket over purple top, and maroon-pink backpack visible beside her. However, hair length changed significantly -- the reference shows medium-length wavy/bob hair, while the new image shows long straight hair. Facial features are reasonably similar but hair is a clear anchor drift.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Same Asian girl with dark hair, teal-blue open jacket, purple top, and pink backpack all present. Hair appears slightly shorter/bobbed here versus the reference which shows medium-length hair, but overall identity anchors hold well.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tunedref gap
continuity4fidelity5style5cross page consistency4

Same Asian girl identity: dark hair, teal-blue open jacket with same circular badge, purple floral top, pink backpack strap visible, tan pants with belt all match. Hair appears slightly shorter/wavier than the reference's shoulder-length bob, but the reference itself doesn't show clearly long hair either, so this is minor drift.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tunedref gap
continuity3fidelity5style5cross page consistency4

Same Asian girl identity with teal-blue open jacket, purple top, and maroon-pink backpack all preserved. However, hair length changed significantly -- the reference shows shoulder-length bobbed hair, while the new illustration shows long straight hair past the shoulders. This is a visible anchor drift from the reference itself.

promptAn illustration with the polished look of a modern 3D animated feature film frame: soft global illumination, warm cinematic color grading, rounded appealing character shapes -- Disney/Pixar 3D animation, warm color grading. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
Vibrant Fantasy

Vibrant Fantasy

Maria Vibrant Fantasy exemplar
reference + style barStable Image Core mintfrozen canonical

Rendered by Stable Image Core from the SIC mint prompt below. This image is BOTH the frozen identity reference for every arm in this section and the visual definition of what the style must look like.

sic mint prompt(SIC prompt not found)

gpt-image-2 -- sic-verbatim

$0.040 / imagecross-page consistency4
library
librarygpt-image-2 · arm: sic-verbatimref gap
continuity5fidelity4style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue open jacket over purple top, matching pendant/necklace detail, same facial features and rosy cheeks. All reference-visible anchors held.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Vibrant fantasy art, jewel-tone palette (analytical-curious:0.6). The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachgpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity4style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue jacket worn open over a purple top, similar facial features and earring. The maroon-pink backpack from the sheet is not visible in the reference either, so not penalized. Minor drift: the jacket looks slightly glossier and the purple top's ornamental buttons are less prominent, but identity anchors hold.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Vibrant fantasy art, jewel-tone palette (analytical-curious:0.6). The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddinggpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity4style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue jacket worn open over a purple top - core anchors preserved. Facial features are recognizably similar. The maroon-pink backpack from the sheet is absent, but it was also absent from the reference, so not a continuity break. Minor drift in facial proportions but recognizably the same character.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Vibrant fantasy art, jewel-tone palette (analytical-curious:0.6). The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardengpt-image-2 · arm: sic-verbatimref gap
continuity4fidelity5style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue open jacket over purple top with gold button details, and similar facial features/earring. Identity anchors from the reference hold well. The maroon-pink backpack from the sheet is not visible, but it was also absent in the reference, so not a continuity penalty.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Vibrant fantasy art, jewel-tone palette (analytical-curious:0.6). The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

gpt-image-2 -- tuned

$0.040 / imagecross-page consistency5
library
librarygpt-image-2 · arm: tunedref gap
continuity5fidelity4style5cross page consistency5

Same girl: long straight black hair, identical facial features, teal-blue jacket worn open over a purple top with gold buttons, earrings preserved. All reference-visible anchors held.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
beach
beachgpt-image-2 · arm: tunedref gap
continuity4fidelity5style5cross page consistency5

Same Asian girl with long straight black hair, teal-blue jacket worn open over a purple top, and gold button/pendant details all preserved. Facial features (rosy cheeks, similar eyes/nose/mouth) match well. The maroon-pink backpack is added here (not visible in reference) but matches the sheet. Minor drift: jacket appears slightly more saturated but essentially the same character.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
sledding
sleddinggpt-image-2 · arm: tunedref gap
continuity4fidelity5style5cross page consistency5

Same Asian girl with long straight black hair, teal-blue jacket worn open over a purple buttoned top -- all key reference-visible anchors preserved. Facial features (round cheeks, dark eyes, small earring) match well. The maroon-pink backpack is newly added per the sheet but was not in the reference, so its presence is consistent with the sheet.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.
garden
gardengpt-image-2 · arm: tunedref gap
continuity5fidelity5style5cross page consistency5

Same Asian girl with long straight black hair, teal-blue jacket worn open over a purple buttoned top with gold buttons, same facial features and rosy cheeks. The maroon-pink backpack is now visible (it wasn't in the reference but is a sheet anchor). All reference-visible anchors are preserved.

promptCharacter: the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack. Scene: planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Style: a richly painted fantasy illustration in a jewel-tone palette of sapphire, emerald, amethyst, and gold, with visible painterly brushwork, luminous decorative flora, and a gouache-on-board texture -- Vibrant fantasy art, jewel-tone palette. No 3D render look, no photorealism, no lettering anywhere in the image.

nano-banana -- sic-verbatim

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: sic-verbatimref gap
continuity4fidelity5style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue jacket open over a purple top, similar facial features and earring. The maroon-pink backpack from the sheet is not visible, but it wasn't visible in the reference either. Minor drift in exact shirt detail but identity anchors from the reference hold.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Vibrant fantasy art, jewel-tone palette (analytical-curious:0.6). The girl is sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby.
beach
beachnano-banana · arm: sic-verbatimref gap
continuity4fidelity4style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue jacket open over purple top -- all reference-visible anchors present. Face shape and features are close, though slightly younger/softer than reference. The maroon-pink backpack from the sheet is not visible, but it was also absent in the reference, so not a continuity penalty.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Vibrant fantasy art, jewel-tone palette (analytical-curious:0.6). The girl is building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand.
sledding
sleddingnano-banana · arm: sic-verbatimref gap
continuity4fidelity4style4cross page consistency4

Same Asian girl with long straight black hair, teal-blue jacket worn open over a purple top -- all reference-visible anchors preserved. Facial features are consistent. Minor drift: the reference shows a more subdued/serious expression while the new image is joyful, but that reflects the scene action, not identity. The maroon-pink backpack from the sheet is absent, but it was also absent from the reference, so not counted against continuity.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Vibrant fantasy art, jewel-tone palette (analytical-curious:0.6). The girl is riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement.
garden
gardennano-banana · arm: sic-verbatimref gap
continuity4fidelity4style4cross page consistency4

Same Asian girl with long straight black hair, teal-blue open jacket over a purple top with gold button/medallion details. Facial features (rosy cheeks, similar eyes/nose) are consistent. The maroon-pink backpack from the sheet is not visible in either image, so not a continuity break. Minor drift: jacket appears slightly more saturated and the purple top styling is simplified, but identity clearly holds.

promptIn a (botanical garden:0.5) a 5th grader Asian girl wearing blue jacket (girl:1.2) in a jewel-tone themed room. Vibrant fantasy art, jewel-tone palette (analytical-curious:0.6). The girl is planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background.

nano-banana -- tuned

$0.039 / imagecross-page consistency4
library
librarynano-banana · arm: tuned
continuity4fidelity5style5cross page consistency4

Same Asian girl with long straight black hair, teal-blue jacket open over a purple top -- key anchors preserved. The maroon-pink backpack is present beside her. Face shape and youthful appearance are consistent, though she looks slightly older/more mature here. Minor drift: reference showed her in a jacket-and-shirt outfit with no visible skirt, while here she wears a long teal skirt/dress, but the top half matches well.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
beach
beachnano-banana · arm: tuned
continuity4fidelity4style2cross page consistency4

Same Asian girl with long straight black hair, teal-blue jacket open over purple top with gold button details preserved. The maroon-pink backpack is present. Facial features are similar. Minor drift in exact face proportions but identity anchors hold well.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
sledding
sleddingnano-banana · arm: tunedref gap
continuity4fidelity5style2cross page consistency4

Long straight black hair, teal-blue open jacket over purple top, and a pink/maroon backpack all carry over from the reference. Facial features read as the same Asian girl, though she appears slightly older/more stylized. Backpack in reference wasn't clearly visible but sheet-anchored; jacket and hair match well.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
garden
gardennano-banana · arm: tunedref gap
continuity4fidelity4style3cross page consistency4

Same Asian girl with long straight black hair, teal-blue jacket open over purple top with gold button details, and earring visible. The maroon-pink backpack is now present (absent in reference). Facial features and rosy cheeks match well. Minor drift: slightly younger/softer face rendering but recognizably the same character.

promptAn illustration with a richly hand-painted fantasy illustration glowing with jewel tones -- sapphire, emerald, amethyst, gold -- built from visible painterly brushwork and luminous decorative flora -- Vibrant fantasy art, jewel-tone palette. It shows the same girl as the attached reference image -- long straight black hair, teal-blue jacket worn open over a purple top, maroon-pink backpack, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background. Match the attached reference image for both the character's identity and the target art style, keeping every surface free of lettering, captions, and panel borders.
Fantasy Art

Fantasy Art

Renders in progress or pending judging -- this section fills in as the arm completes.

Full written results

the complete document behind this page -- method, every table, the reads, and the standing caveats

_Last updated: 2026-07-30_

Experiment 3 -- style adherence, control vs treatment: running results

The round exists because judge scores and the eye disagreed: experiment 2 style arms scored fine, but the rendered graphic-novel look comes back as cheap flat cartoon instead of comic-book inking. This round tests whether prompt design closes that gap, per style, per model, with the delta between two arms as the measurement.

Status: all 96 renders dispatched 2026-07-30 (3 styles x 4 models x 2 arms x 4 scenes, character maria-v2, style-matched SIC exemplars as references). Judging is opus-over-Bedrock and queued behind an AWS credential refresh; sections below fill in per style as verdicts land, gallery updated after each.

Design

  • Character: maria-v2 -- the new Maria design from the three SIC exemplars (fifth-grade Asian girl, long straight black hair, teal-blue jacket over purple top, maroon-pink backpack). Sheet carries only exemplar-stable anchors; the backpack is absent from the vibrant-fantasy exemplar, so that arm treats it as reference_gap, not drift.
  • References: the SIC exemplars themselves, frozen as references/maria-v2-&lt;style&gt;.png. Each is simultaneously the identity reference and the style bar. The SIC mint prompts (verbatim, weighted syntax preserved) are recorded in the gallery on each exemplar card and in utilities/SIC_prompts_by_image.md.
  • Arm A, sic-verbatim (control): the exact SIC mint prompt, unmodified, plus the scene clause appended identically for every model: &lt;SIC prompt&gt;. The girl is &lt;scene clause&gt;. This measures shipping the current production prompt idiom to these models as-is -- weighted (term:weight) syntax included, which neither API documents support for.
  • Arm B, tuned (treatment): per-family idiom from the prompting playbooks. gpt family: labeled segments (Character: &lt;lock&gt;. Scene: &lt;clause&gt;. Style: &lt;properties&gt; -- &lt;picker string&gt;. &lt;exclusions&gt;), exclusions naming the cheap-cartoon failure directly. Gemini: style-first narrative sentences, semantic positives only, one text-suppression sentence, no exclusion lists. Both arms keep the verbatim picker string. Identity uses the short lock over the reference (E1/N1 direction), not the full sheet. Exact strings: lib/style-prompt-variants.mjs in the harness; every record also carries its prompt verbatim.
  • Models: gpt-image-1, gpt-image-1-mini, gpt-image-2 (new registry entries for the latter two; -2 omits input_fidelity by contract), nano-banana.
  • Cost: ~$1 per style across the four models at medium quality, ~$3 for the round, plus judge tokens.

Research salvage (the cursor session's CME agents were cut off by usage credits; verified findings kept)

  • Gemini 2.5-flash-image has NO style-reference input: style-reference images (up to 3) are documented only for Gemini 3 Pro Image. On 2.5, text carries the style; our style-matched identity reference helps informally, but nothing in the API contract makes it a style carrier. That puts the whole style signal on the tuned arm's opening sentences for nano.
  • Google's own cookbook pattern for style-carrying on this family is a style-first prefix (establish the style, then the subject) -- the tuned nano prompts follow it.
  • Brand-risk precedent on Gemini: press-documented rejections of Disney-owned-property image requests. Our own "Disney/Pixar" renders have passed (8+ with zero blocks, plus this experiment's 3d-animation arms), but the string stays a flagged hazard with a studio-free fallback ready, not a cleared item.

Known incidents this experiment

  • One transient render failure on the first call of the round (gpt-image-1, graphic-novel, sic-verbatim, library scene): HTTP 200 with an unparseable body. A one-retry guard for exactly this class is now in openai-image.mjs (mirroring the Gemini chatty-no-image guard); the missing cell gets patched by a single re-render before judging.

Results: graphic-novel (judged 2026-07-30, opus)

ModelA: sic-verbatimB: tunedDeltaA styleB styleB consistency
gpt-image-13.754.00+0.254.755.004
gpt-image-1-mini4.003.75-0.255.005.004
gpt-image-24.255.00+0.755.005.005
nano-banana4.005.00+1.004.754.505

Continuity columns are the 4-scene average; style is the judge's adherence-to-requested-style average.

The headline: gpt-image-2 with the tuned prompt is the first perfect arm in the entire evaluation -- 5.00 continuity, 5.00 style, 5 cross-page consistency, every scene. Nano-banana's tuned arm also hit 5.00 continuity; prompt tuning bought it a full point over the control.

The eye test, which is why this experiment exists: the tuned renders come back as actual comic-book interior art -- varied ink line weights, hard-edged cel shadows, saturated flat fills -- on both gpt-image-2 and nano-banana. gpt-image-2's tuned pages sit visually next to the SIC exemplar; nano's line is cleaner and lighter but unmistakably inked comic style, not the flat basic cartoon of experiment 2. The property-language approach (naming ink, cel shadows, and fills instead of trusting the words "graphic novel") plus the exclusion clause on the gpt side is what changed.

The control arm is the migration-risk read: the raw production prompt, weighted syntax and all, still scored 3.75-4.25 continuity with style around 4.75-5.0 -- serviceable, no catastrophic breakage, and no visible weight-syntax text leakage on any model (the judge's notes never flag stray text; the nano hazard did not materialize this style). So shipping today's prompt idiom unchanged would not fall over, but it leaves 0.75-1.00 continuity on the table for the two models that matter.

Odd result worth keeping: gpt-image-1-mini is prompt-indifferent (the only negative delta, -0.25, within noise at n=4) -- consistent with its draft-tier character: it renders what it renders regardless of prompt sophistication.

Results: 3d-animation (judged 2026-07-30, opus)

ModelA: sic-verbatimB: tunedDeltaA styleB style
gpt-image-13.003.50+0.504.505.00
gpt-image-1-mini3.003.50+0.504.505.00
gpt-image-24.003.25-0.755.005.00
nano-banana4.003.50-0.504.754.75

The tuning delta REVERSED on this style for the two strongest models. On graphic-novel, tuned prompts bought gpt-image-2 and nano-banana +0.75 and +1.00; here they cost them -0.75 and -0.50, while still helping the two weaker models by +0.50. Style adherence sat at 4.5-5.0 in every arm either way.

Working hypothesis, consistent with the prompting research: the tuned 3d block is heavy with rendering-property language ("glossy feature film still, subsurface-scattered skin, depth of field"), which pushes an aggressive re-render that competes with the reference for identity. On an inked style that language sharpened linework without touching identity; on a photoreal-adjacent rendered style it invites the model to re-imagine the character. The control prompt says almost nothing stylistic beyond the picker string, so the style-matched reference dominates -- and the reference IS the identity. Lesson shaping up: tuned property language earns its keep on styles the model renders poorly by default, and should stay minimal on styles adjacent to the model's native look. n=4 per cell; treat directions, not decimals.

Also notable: continuity runs lower across the board on this style (3.00-4.00) than on graphic-novel, with reference_gap flags on most arms -- the 3d exemplar hides several sheet anchors (backpack barely visible, top mostly covered), so the judge is working from a thinner anchor set. Reference framing matters again.

The "Disney/Pixar" string cleared another 32 renders across four models and two arms with zero moderation blocks -- the largest single-day exposure yet.

Results: vibrant-fantasy (judged 2026-07-30, opus)

ModelA: sic-verbatimB: tunedCont deltaA styleB styleStyle delta
gpt-image-14.004.000.004.254.00-0.25
gpt-image-1-mini3.754.00+0.254.505.00+0.50
gpt-image-24.254.50+0.255.005.000 (consistency 5)
nano-banana4.004.000.004.503.00-1.50

Mostly quiet deltas -- gpt-image-2 tuned again the best arm (4.50 continuity, 5.00 style, consistency 5) -- except one loud failure: nano-banana's tuned arm collapsed on style adherence (3.00 vs its own control's 4.50). The judge's notes are consistent scene to scene: the renders drifted into "soft pastel children's-book illustration," losing the jewel-tone painterly brief on 3 of 4 scenes. The terse control prompt held the style; the long style-first narrative did not. Reading this with the 3d result: nano's drift-to-soft-attractor is scene-content-sensitive, and long passages appear to DILUTE its style signal rather than reinforce it (consistent with the documented attention-dilution weakness when a prompt asks for many things at once). Nano wants short, dense style language, not rich narrative.

The read

Across three styles, four models, and both arms (96 renders, opus-judged, style-matched SIC exemplar references throughout):

  1. gpt-image-2 tuned is the best arm of the round on every style -- 5.00/5.00/5 on graphic-novel (the first perfect arm in the evaluation), 4.50/5.00/5 on vibrant-fantasy, and even where tuning cost it continuity (3d-animation) it stayed at 5.00 style. If one model-prompt pairing ships tomorrow, it is gpt-image-2 with per-style-calibrated prompts.
  2. The comic-book complaint is answered. Tuned property-language prompts moved graphic-novel from cheap flat cartoon to genuine comic interior art -- confirmed by the judge (5.00 style on three of four models) AND by eye against the exemplar. This was the round's reason to exist, and it is a yes: prompt tuning covers the difference.
  3. But tuning is style-dependent, not universal. Deltas by style: graphic-novel +0.25/-0.25/+0.75/+1.00, 3d-animation +0.50/+0.50/-0.75/-0.50, vibrant-fantasy 0/+0.25/+0.25/0 with the nano style collapse. The pattern: property language pays where the target style is far from the model's default (inked comic); it fights the reference where the style is near the default (3d), and on nano long narrative dilutes rather than reinforces. Production implication: the style prompt table should be per-model AND per-style, short where short wins -- not one enriched template.
  4. The control arm says a naive migration would not fall over. The raw production SIC prompt, weighted syntax and all, scored 3.75-4.25 continuity everywhere with no weight-syntax text leakage on any model, any style. Serviceable -- but it leaves up to a full point of continuity and the entire comic-book jump on the table.
  5. The "Disney/Pixar" string is now 60+ renders across four models with zero moderation blocks. Still not a policy guarantee, but the empirical block rate to date is 0.
  6. Caveats: n=4 per cell (directions, not decimals); single character; the 3d exemplar hides several sheet anchors (ref-gap flags throughout that style, and its lower continuity partly reflects reference framing); mini remains prompt-indifferent within noise in either direction.

Next moves this sets up: a per-style prompt-profile table in the harness (the variants module already supports it), a nano-specific short-form style retry on vibrant-fantasy, and the E1/N1 identity-slot A/B, which is now the last big untested lever.

Story Studio image styles -- model tests

Findings
  • With style-matched references the two models essentially tie -- reference quality drives continuity more than model choice on this slice.
  • Style adherence flips to a win when the reference matches the target style (gpt 5.0 on vibrant-fantasy, nano 5.0 on 3D -- the very style nano drifted toward with mismatched refs).
  • Fantasy-illustration lags on both models, exactly where the references are weakest (missing anchors, ~200px crop).
Recommended next steps
  • QC the reference mint before debating models (this became experiments 3/3b's reference-strength finding).
  • Test prompt arms per model per style -- executed as experiments 3/3b, one tab over.
Read the full written results for this experiment ↓method, every table, the reads, and the standing caveats

One section per picker style, the picker's own thumbnail as the visual target. Within each style: the characters tested, and under each character the two models side by side -- same frozen reference, byte-identical prompts, scored by the opus judge on continuity, scene fidelity, and adherence to the requested style.

judge claude-opus-4-7 via Bedrock on every scored cardgraphic-novel continuity gpt 3.92 / nano 4.00vibrant-fantasy (Zuri) 4.25 / 4.25, style adherence 4.75 / 4.00runs 2026-07-22 to -28
Vibrant Fantasy target

Vibrant Fantasy

picker string, verbatim in every prompt: see each card

Zuri

human childvoluminous curly hair, round black glasses, teal puffer vest, graphic tee, jeans

nano-banana

cross-page consistency4
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference · style=vibrant-fantasy
continuity5fidelity5style4cross page consistency4

All reference-visible anchors preserved: brown skin, voluminous dark curly hair, round black-rimmed glasses, teal puffer vest open over cream graphic tee, blue jeans, and sneakers (teal accents). Same character clearly.

promptZuri, a young girl with brown skin and voluminous dark curly hair, round black-rimmed glasses, a bright teal puffer vest worn open over a cream graphic t-shirt, blue jeans, and sneakers, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, Vibrant fantasy art, jewel-tone palette
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=vibrant-fantasy
continuity4fidelity5style4cross page consistency4

Same brown-skinned child with voluminous dark curly hair, round black-rimmed glasses, teal puffer vest over cream tee, blue jeans, and sneakers -- all reference anchors present. Minor drift: face shape looks slightly younger/rounder and vest styling is a bit different, but identity clearly holds.

promptZuri, a young girl with brown skin and voluminous dark curly hair, round black-rimmed glasses, a bright teal puffer vest worn open over a cream graphic t-shirt, blue jeans, and sneakers, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, Vibrant fantasy art, jewel-tone palette
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=vibrant-fantasy
continuity4fidelity5style4cross page consistency4

Same character: brown skin, voluminous dark curly hair, round black-rimmed glasses, teal puffer vest open over cream graphic t-shirt, blue jeans all preserved. Sneakers are replaced with winter boots, which is scene-appropriate but a minor clothing drift. A pink hair accessory is added that wasn't in the reference.

promptZuri, a young girl with brown skin and voluminous dark curly hair, round black-rimmed glasses, a bright teal puffer vest worn open over a cream graphic t-shirt, blue jeans, and sneakers, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, Vibrant fantasy art, jewel-tone palette
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference · style=vibrant-fantasy
continuity4fidelity4style4cross page consistency4

Same character is clearly recognizable: brown skin, voluminous dark curly hair, round black-rimmed glasses, teal puffer vest open over cream graphic t-shirt, blue jeans. Sneakers are now teal/green rather than an unspecified color in the reference (hard to tell in reference), a minor drift but overall identity holds.

promptZuri, a young girl with brown skin and voluminous dark curly hair, round black-rimmed glasses, a bright teal puffer vest worn open over a cream graphic t-shirt, blue jeans, and sneakers, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, Vibrant fantasy art, jewel-tone palette

Maria

human childdark shoulder-length hair with headband, blue jacket over warm orange layer
Maria Vibrant Fantasy reference
referenceStable Image Core exemplar · style-matchedfrozen canonical

Rendered by Stable Image Core as the style exemplar for this arm (SIC is the style mint, not a model under test). Continuity below is judged against THIS image.

nano-banana

cross-page consistency3
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=vibrant-fantasy
continuity5fidelity5style4cross page consistency3

Same girl: light-brown skin, dark shoulder-length hair, orange headband, teal/blue jacket with magenta hood lining, orange bow/scarf, and orange inner layer all preserved. Facial features closely match the reference.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, Vibrant fantasy art, jewel-tone palette
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=vibrant-fantasy
continuity4fidelity5style4cross page consistency3

Same girl: light-brown skin, dark shoulder-length hair, orange headband, teal/blue jacket with magenta cuffs, orange inner layer with orange bow/scarf, dark pants, brown boots. All key reference-visible anchors are preserved. Face reads slightly younger/more cheerful but identity holds.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, Vibrant fantasy art, jewel-tone palette
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference · style=vibrant-fantasyref gap
continuity3fidelity4style3cross page consistency3

Same girl archetype with light-brown skin, dark shoulder-length hair, headband, blue jacket, and orange inner layer -- all core anchors present. However, the headband color shifted from red/magenta in reference to orange, and the jacket changed from a teal/purple-accented coat with orange scarf to a plain blue hoodie. Facial features also read younger and softer. Recognizably related but drifted.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, Vibrant fantasy art, jewel-tone palette
3D Animation target

3D Animation

picker string, verbatim in every prompt: see each card

Maria

human childdark shoulder-length hair with headband, blue jacket over warm orange layer
Maria 3D Animation reference
referenceStable Image Core exemplar · style-matchedfrozen canonical

Rendered by Stable Image Core as the style exemplar for this arm (SIC is the style mint, not a model under test). Continuity below is judged against THIS image.

nano-banana

cross-page consistency4
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference · style=3d-animation
continuity4fidelity5style5cross page consistency4

Same girl: light-brown skin, dark shoulder-length hair, red/dark headband, blue hooded jacket over orange inner layer all preserved. Face shape and expression are consistent. Minor drift: the skirt from the reference isn't clearly visible (she has a blanket/book over her lap), but leggings and boots match.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, Disney/Pixar 3D animation, warm color grading
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=3d-animation
continuity3fidelity5style5cross page consistency4

Same girl: light-brown skin, dark shoulder-length hair, red headband, similar face shape. However, the blue jacket and warm orange inner layer are missing -- she's wearing only a short-sleeved blue t-shirt. The jacket/orange layering is a key reference-visible anchor that's lost.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, Disney/Pixar 3D animation, warm color grading
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=3d-animation
continuity3fidelity5style5cross page consistency4

Same light-brown skin, dark hair, blue jacket over orange inner layer, and dark leggings are preserved. However, the headband -- a clear reference anchor -- is missing in the new illustration. Face shape and hair also read slightly older/longer, but identity is still recognizable.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, Disney/Pixar 3D animation, warm color grading
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference · style=3d-animation
continuity4fidelity4style5cross page consistency4

Same girl: light-brown skin, dark shoulder-length hair, red headband, blue jacket all preserved. The warm orange inner layer is barely visible (just a hint at the wrist/collar), and the skirt appears to have shifted to pants/tan pants, but core identity anchors hold.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, Disney/Pixar 3D animation, warm color grading
Fantasy Illustration target

Fantasy Illustration

picker string, verbatim in every prompt: see each card

Maria

human childdark shoulder-length hair with headband, blue jacket over warm orange layer
Maria Fantasy Illustration reference
referenceStable Image Core exemplar · style-matchedfrozen canonical

Rendered by Stable Image Core as the style exemplar for this arm (SIC is the style mint, not a model under test). Continuity below is judged against THIS image.

nano-banana

cross-page consistency3
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference · style=fantasy-illustrationref gap
continuity2fidelity5style3cross page consistency3

The reference shows an East Asian girl with straight black hair, pale skin, and a blue hooded coat over a pink/magenta inner layer with an orange neckerchief -- no headband is visible. The new illustration shows a girl with brown (not black) hair, an orange headband, lighter/warmer skin tone, and a blue hoodie over an orange inner layer. Facial features, hair color, and overall ethnicity read differently -- this looks like a generic look-alike rather than the same child.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, Hand-painted watercolor storybook illustration
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=fantasy-illustrationref gap
continuity3fidelity5style5cross page consistency3

Recognizable as a young girl with dark hair and blue jacket over orange layer, matching sheet anchors. However, the reference showed no headband and long straight hair with an East Asian appearance, while the new image shows a colorful headband (matches sheet, not reference), shorter wavier hair, and different facial features. The identity has drifted noticeably from the reference.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, Hand-painted watercolor storybook illustration
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=fantasy-illustrationref gap
continuity3fidelity5style5cross page consistency3

Same young girl with light skin and dark shoulder-length hair is recognizable. However, the headband is now visible (it wasn't clearly in the reference), the orange inner layer has become an orange neckerchief/scarf (matching reference), and the blue jacket is present but simplified. Pants changed from beige to brown. Overall recognizably related but some anchors drift.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, Hand-painted watercolor storybook illustration
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference · style=fantasy-illustrationref gap
continuity3fidelity4style5cross page consistency3

Same young girl with dark shoulder-length hair and light skin, wearing blue jacket over orange inner layer -- these anchors hold. However, the reference does NOT show a headband (hair is loose), and the new image adds an orange headband. The face shape is similar but slightly older-looking. Core anchors visible in reference (hair, skin, blue jacket, orange layer) are preserved, though the added headband is a drift.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, Hand-painted watercolor storybook illustration

Mr Penguin

penguinblack back, white belly, black chest band, white face stripe, orange beak and feet
Mr Penguin Fantasy Illustration reference
referenceStable Image Core exemplar · style-matchedfrozen canonical

Rendered by Stable Image Core as the style exemplar for this arm (SIC is the style mint, not a model under test). Continuity below is judged against THIS image.

nano-banana

cross-page consistency3
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference · style=fantasy-illustrationref gap
continuity3fidelity5style5cross page consistency3

The new illustration shows a penguin with black back and white belly, but the reference clearly displays a Magellanic-style black chest band and white facial stripe curving around the face, plus a black/dark beak. The new image shows a rounder, more cartoonish penguin with an orange beak, no visible chest band, and no curved white face stripe -- key reference-visible anchors are missing. Orange feet are present.

promptMr Penguin, a Magellanic penguin with a black back, white belly, a black band across the chest, a white stripe curving around the face, and an orange beak and orange webbed feet, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, Hand-painted watercolor storybook illustration
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=fantasy-illustration
continuity3fidelity5style5cross page consistency3

Same penguin species with black back, white belly, orange beak and feet. However, the reference shows a clear black chest band across the white belly, which is missing/faint in the new illustration (belly is mostly plain white with a small yellow patch). The white facial stripe is also less pronounced. Recognizable as Mr Penguin but a key anchor (chest band) drifted.

promptMr Penguin, a Magellanic penguin with a black back, white belly, a black band across the chest, a white stripe curving around the face, and an orange beak and orange webbed feet, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, Hand-painted watercolor storybook illustration
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=fantasy-illustrationref gap
continuity2fidelity5style5cross page consistency3

Reference shows a Gentoo-like penguin with black head, white belly, orange beak and feet, but no black chest band and no curving white face stripe visible either. The new illustration adds a yellow/cream chest patch (emperor-like), lacks the white face stripe, and shows a rounder chick-like body proportion. The orange beak and feet carry over, but the head/face pattern and body shape differ notably, making it feel like a different penguin species rather than the same character.

promptMr Penguin, a Magellanic penguin with a black back, white belly, a black band across the chest, a white stripe curving around the face, and an orange beak and orange webbed feet, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, Hand-painted watercolor storybook illustration
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference · style=fantasy-illustrationref gap
continuity3fidelity5style5cross page consistency3

Species (penguin), black back, white belly, orange beak, and orange feet are preserved. However, the distinctive black chest band and the white facial stripe curving around the face from the reference are largely lost -- the new illustration shows a different face pattern (white patch around the eye/cheek area) and no clear chest band, giving a slightly different Magellanic identity.

promptMr Penguin, a Magellanic penguin with a black back, white belly, a black band across the chest, a white stripe curving around the face, and an orange beak and orange webbed feet, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, Hand-painted watercolor storybook illustration
Graphic Novel target

Graphic Novel

picker string, verbatim in every prompt: see each card

Maya

human childcurly hair puffs, round red glasses, yellow raincoat, green boots

nano-banana

cross-page consistency4
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity5fidelity5style4cross page consistency4

All reference-visible anchors are preserved: brown skin, curly black hair in two puffs, round red glasses, open bright yellow raincoat over blue-and-white striped shirt, and green rubber boots. Facial proportions and features read as the same character.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, Graphic novel style, bold ink outlines, cel-shaded
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity4fidelity5style5cross page consistency4

Same character clearly recognizable: brown skin, two curly black hair puffs, round red glasses, open yellow raincoat over blue/white striped shirt, and green rubber boots all present. Minor drift in facial expression (more neutral/concerned vs. smiling) but all identity anchors survive.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, Graphic novel style, bold ink outlines, cel-shaded
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity5fidelity5style5cross page consistency4

All reference-visible anchors preserved: brown skin, two curly black hair puffs, round red glasses, open bright yellow raincoat over blue/white striped shirt, blue shorts, and green rubber boots. Identity is unmistakably Maya.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, Graphic novel style, bold ink outlines, cel-shaded
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity5fidelity4style5cross page consistency4

All reference-visible anchors are preserved: brown skin, curly black hair in two puffs, round red glasses, open yellow raincoat over blue-and-white striped shirt, and green rubber boots (one visible). Facial features and proportions read as the same character.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, Graphic novel style, bold ink outlines, cel-shaded

Pepper

raccoonnotched left ear, star chest patch, green backpack

nano-banana

cross-page consistency3
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novelref gap
continuity3fidelity5style4cross page consistency3

Same gray raccoon with black mask, ringed tail, green backpack, and similar smile. However, the white star-shaped chest patch is not clearly visible (obscured or missing), and no notch in the left ear is visible - though the notch was also not visible in the reference. The chest star being absent is a notable anchor loss.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, Graphic novel style, bold ink outlines, cel-shaded
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novelref gap
continuity3fidelity5style5cross page consistency3

Same species, gray fur, black mask, ringed tail, and green backpack all carry over. However, the white star-shaped chest patch is missing/reduced to a faint mark on the backpack area, and the left ear notch is not visible -- both key reference-visible anchors are absent or unclear.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, Graphic novel style, bold ink outlines, cel-shaded
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novelref gap
continuity4fidelity5style5cross page consistency3

Same gray raccoon with black mask, ringed tail, white star chest patch, and green backpack all preserved. The notch in the left ear is not visible in the reference either, so not penalized. Minor drift: proportions are slightly chubbier/rounder in the new illustration, but identity anchors hold.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, Graphic novel style, bold ink outlines, cel-shaded
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novelref gap
continuity4fidelity4style5cross page consistency3

Same gray raccoon with black mask, ringed tail, white star chest patch, and green backpack all preserved. The star has 6 points vs the reference's rounder star shape, a minor drift. Ear notch is not clearly visible in reference so not penalized.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, Graphic novel style, bold ink outlines, cel-shaded

Bolt

robotone big eye, star antenna, tank treads, red belly hatch

nano-banana

cross-page consistency4
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity3fidelity5style4cross page consistency4

Teal color, one large eye, star-tipped antenna, and rounded body are preserved. However, the tank treads are largely hidden/replaced with what look like simple feet under the chair, and the red belly hatch is not visible. Two key reference-visible anchors are missing or obscured.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, Graphic novel style, bold ink outlines, cel-shaded
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity4fidelity5style5cross page consistency4

Same teal round robot with one large friendly eye, star-topped antenna, tank treads, and red belly hatch all present. Minor drift: body proportions appear slightly more elongated/less chunky than reference, but identity anchors hold.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, Graphic novel style, bold ink outlines, cel-shaded
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity4fidelity5style5cross page consistency4

Teal round body, single large eye, star-topped antenna, red belly hatch, and tank treads are all present. Minor drift: the treads appear as two separate wheel/tread units rather than the reference's stubby dual-wheel legs, and proportions are slightly slimmer, but identity is unmistakable.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, Graphic novel style, bold ink outlines, cel-shaded
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity4fidelity5style5cross page consistency4

Same character clearly recognizable: teal round body, single large friendly eye, antenna with yellow star, red hatch on belly all present. Tank treads are barely visible/mostly hidden behind the garden bed but a hint of a wheel is shown. Proportions and color hold well.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, Graphic novel style, bold ink outlines, cel-shaded

Maria

human childdark shoulder-length hair with headband, blue jacket over warm orange layer
Maria Graphic Novel reference
referenceStable Image Core exemplar · style-matchedfrozen canonical

Rendered by Stable Image Core as the style exemplar for this arm (SIC is the style mint, not a model under test). Continuity below is judged against THIS image.

nano-banana

cross-page consistency4
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity4fidelity5style4cross page consistency4

Same girl: light-brown skin, dark shoulder-length hair, yellow headband, blue jacket with orange inner layer all preserved. Face shape and features read as the same character. Minor drift: expression is softer and the jacket looks more like a cardigan/sweater than the structured hooded jacket in the reference, but identity anchors hold.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, Graphic novel style, bold ink outlines, cel-shaded
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity4fidelity5style4cross page consistency4

Same girl: light-brown skin, dark shoulder-length hair, yellow headband, blue jacket with orange inner layer, and backpack all preserved. Facial features are consistent. Minor drift: the orange layer reads more as a hood/collar than a full inner layer, and hair appears slightly shorter, but overall unmistakably Maria.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, Graphic novel style, bold ink outlines, cel-shaded
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity4fidelity5style4cross page consistency4

Same girl: light-brown skin, dark shoulder-length hair, yellow headband, blue jacket over orange inner layer all preserved. Backpack carried over. Face is slightly rounder/younger looking than the reference, a minor drift, but identity anchors hold.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, Graphic novel style, bold ink outlines, cel-shaded
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity4fidelity4style4cross page consistency4

Same girl: light-brown skin, dark shoulder-length hair, yellow headband, blue jacket with warm orange inner hood layer, and white/navy shirt underneath all match the reference. Facial features are consistent though slightly softened. One minor drift: her expression and eye shape are a bit different, but identity anchors hold.

promptMaria, a young girl with light-brown skin and dark shoulder-length hair held by a headband, wearing a blue jacket over a warm orange inner layer, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, Graphic novel style, bold ink outlines, cel-shaded

Mr Penguin

penguinblack back, white belly, black chest band, white face stripe, orange beak and feet
Mr Penguin Graphic Novel reference
referenceStable Image Core exemplar · style-matchedfrozen canonical

Rendered by Stable Image Core as the style exemplar for this arm (SIC is the style mint, not a model under test). Continuity below is judged against THIS image.

nano-banana

cross-page consistency3
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity4fidelity5style5cross page consistency3

Same Magellanic penguin identity: black back, white belly, black chest band, white curving facial stripe, orange beak, and orange webbed feet all present. Minor drift: the reference shows a single clean chest band, while the new image shows a doubled/thicker band arrangement, but overall clearly the same character.

promptMr Penguin, a Magellanic penguin with a black back, white belly, a black band across the chest, a white stripe curving around the face, and an orange beak and orange webbed feet, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, Graphic novel style, bold ink outlines, cel-shaded
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity4fidelity5style5cross page consistency3

Same penguin identity: black back, white belly, white facial stripe curving around the face, orange beak, orange webbed feet all preserved. The reference shows a subtle darker band suggesting the chest band, which is less visible in the new illustration, but overall the character is unmistakably the same.

promptMr Penguin, a Magellanic penguin with a black back, white belly, a black band across the chest, a white stripe curving around the face, and an orange beak and orange webbed feet, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, Graphic novel style, bold ink outlines, cel-shaded
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity3fidelity5style4cross page consistency3

Same species and core colors (black back, white belly, orange beak/feet) are preserved. However, the distinguishing white stripe curving around the face and the black chest band visible in the reference are missing in the new illustration -- the face is solid black without the white curve, making it look more like a generic/emperor-style penguin than the reference's Magellanic markings.

promptMr Penguin, a Magellanic penguin with a black back, white belly, a black band across the chest, a white stripe curving around the face, and an orange beak and orange webbed feet, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, Graphic novel style, bold ink outlines, cel-shaded
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference · style=graphic-novel
continuity3fidelity5style5cross page consistency3

Same species and overall coloring (black back, white belly, white facial stripe, orange beak) are preserved. However, the reference clearly shows orange webbed feet, and in the new illustration the feet are hidden/obscured and the flippers appear as white/gloved hands rather than the black flippers seen in the reference. The chest band is also less distinct.

promptMr Penguin, a Magellanic penguin with a black back, white belly, a black band across the chest, a white stripe curving around the face, and an orange beak and orange webbed feet, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, Graphic novel style, bold ink outlines, cel-shaded

Full written results

the complete document behind this page -- method, every table, the reads, and the standing caveats

_Last updated: 2026-07-27_

Nano Banana battery results + the opus-judged 2x2

Nano Banana (gemini-2.5-flash-image, Gemini API on a billed AI Studio key) ran the full battery on 2026-07-27: same frozen references, byte-identical scene prompts, baseline arm plus the graphic-novel style arm. Every arm in the table below is scored by the same opus-over-Bedrock judge as the pass-2 baseline, after the test-account credentials fix let the queued re-judges run. Raw runs are in the harness under the run archive (2026-07-27 dirs); probe images under the run archive.

The 2x2 (character continuity, 1-5, n=12 per cell)

Default styleGraphic-novelStyle cost
gpt-image-1 (medium)4.423.92-0.50
nano-banana4.004.000.00

Cross-page consistency: gpt-image-1 4.33 / 4.00; nano-banana 4.00 / 3.67.

Per character, baseline (gpt-image-1 vs nano-banana): Maya 5.00 vs 4.25, Pepper 4.00 vs 3.75, Bolt 4.25 vs 4.00. gpt-image-1 wins or ties every character at baseline; nano-banana's graphic-novel arm actually scored its Maya HIGHER than its own baseline (4.75 vs 4.25) -- at n=4 treat that as noise, but the direction supports the style-robustness read.

What the numbers say

  1. gpt-image-1 is the identity-precision leader in default style (4.42 vs 4.00). Opus's nano dings are the same anchor classes we know: Pepper's star patch degraded (library scored 3 -- the probe render had already shown the patch vanishing), small-mark crispness, one Bolt library 3.
  2. Nano Banana holds its style score where gpt-image-1 does not. The graphic-novel swap cost gpt-image-1 half a point (4.42 to 3.92); nano-banana held flat (4.00 to 4.00). If the style picker ships, this difference matters as much as the baseline gap.
  3. The judge gap is real and now measured twice. GPT-4o scored the gpt-image-1 graphic-novel arm 4.92 (opus: 3.92) and nano-banana ~5.0 in both arms (opus: 4.00). Same pixels, one full point. No cross-judge number appears in this doc's comparisons; the leniency lesson is written up as an article draft (rejudge-image-model-results, 2026-07-27) and baked into the harness rule: comparisons only within one judge column.
  4. Operationally nano-banana is 3-4x faster (~8s/render vs ~25-40s) at essentially the same price ($0.039 vs $0.042 medium), and its scenes read richer/more atmospheric -- scene fidelity 4.67-4.75 with lush detail. It also likes rendering text unprompted ("MAYAY" on a book cover in probe 5d); watch that as a failure family if it ships.

The access + policy facts this run established

  • A personal Google AI Studio key CAN drive this model -- but only with billing linked: the API free tier has ZERO image-generation allowance (probe returned 429 limit: 0; fixed by prepay + project link).
  • A human child character rendered from a reference with no safety block (probe 5d, Maya), repeated across the full battery -- 32 child-character renders total, zero blocks. This is the first empirical data against the Google person/child policy risk. It is a data point, not a policy clearance: the written Google-rep confirmation remains the production gate, and a personal key's terms are not the company's.
  • Quota metrics internally resolve the stable id to gemini-2.5-flash-preview-image; requests with the stable id work. Worth re-checking at contract time given the unconfirmed 2026-10-02 retirement rumor for this model id.

Where this leaves the recommendation

The pass-2 recommendation stands with one amendment. gpt-image-2 remains the next battery (expected to inherit the family's identity precision; its automatic high-fidelity input may also blunt the style-change weakness -- or worsen it, per the research pass's transformation-resistance note; the E3 slice now explicitly includes the style arm to find out). Nano Banana graduates from "frontrunner pending policy" to a measured second: baseline continuity 0.42 behind on the stricter judge, steadier across styles, faster, price-equal, and now empirically past the child-render question on a test key. If gpt-image-2's style arm degrades like gpt-image-1's did, the style-robustness column may decide the whole comparison.

Still open: the E1-E5 experiments backlog (identity-slot A/B could lift EITHER model's scores -- the 4.42 and 4.00 were both earned with the full-sheet prompt the research pass flagged as a possible drift vector), the benign-block battery, and Nano Banana 2 (gemini-3.1-flash-image) as the successor if the retirement rumor firms up.

Story Studio character continuity -- the pictures

Findings
  • Sending a reference image along with the prompt does hold the character. gpt-image-2 averages 4.58 out of 5 on identity at the default style and 4.42 on graphic novel, re-run on 2026-08-07 after the older gpt-image-1 tier was retired. nano-banana averages 4.00 and holds style better of the two. Nova Canvas never entered: our account is locked out of it, and its reference modes carry style and layout rather than identity anyway.
  • Thumbnail-grade references are viable (Zuri's ~128px picker crop tied at 4.25 continuity) -- but weak references cap every downstream score.
  • Judge design matters: continuity is scored against the reference image, with sheet-vs-reference mismatches quarantined as ref gaps.
Recommended next steps
  • Move the battery to gpt-image-2 -- done; experiments 3 onward run on it, and it is the standing ship recommendation.
  • Still open: the benign-block moderation battery, and written Google confirmation on child-content policy before any Gemini production use.
Read the full written results for this experiment ↓method, every table, the reads, and the standing caveats

The baseline batteries: each character's frozen reference and both models' default-style renders, scored by the opus judge, plus the input-fidelity arm and the access probes. Style-arm renders are on the front page.

judge claude-opus-4-7 via Bedrockbaseline continuity gpt-image-2 4.58 / nano-banana 4.00runs 2026-07-22 to 08-07

The opus-judged 2x2

character continuity, n=12 per cell, same judge everywhere -- the only numbers on this page that may be compared to each other
default stylegraphic novelstyle cost
gpt-image-24.584.42-0.16
nano-banana4.004.000.00

Maya

human childcurly hair puffs, round red glasses, yellow raincoat, green boots
Maya reference
referenceminted once · medium · /generationsfrozen canonical

Minted once, then reused for every scene, every model, every future run. Continuity is judged against this image.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, standing in a relaxed friendly pose, plain soft cream background, full body visible, children's book illustration style, warm colors, simple shapes

gpt-image-2

cross-page consistency5
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity5fidelity5style5cross page consistency5

All reference-visible anchors preserved: brown skin, curly black hair in two puffs, round red glasses, open yellow raincoat over blue-and-white striped shirt, and green rubber boots. Same friendly face and proportions.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, children's book illustration style, warm colors, simple shapes
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity5fidelity5style5cross page consistency5

All reference-visible anchors are present: brown skin, black hair in two puffs, round red glasses, open yellow raincoat over blue/white striped shirt, blue shorts, and green rubber boots. Same friendly expression and proportions.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, children's book illustration style, warm colors, simple shapes
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity5fidelity5style5cross page consistency5

All reference-visible anchors are preserved: brown skin, black curly hair in two puffs, round red glasses, open yellow raincoat over blue-and-white striped shirt, blue shorts, and green rubber boots. Facial features and proportions match.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, children's book illustration style, warm colors, simple shapes
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity5fidelity5style5cross page consistency5

All reference-visible anchors are preserved: brown skin, black curly hair in two puffs, round red glasses, open yellow raincoat over blue-and-white striped shirt, and green rubber boots visible on the kneeling figure.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, children's book illustration style, warm colors, simple shapes

nano-banana

cross-page consistency4
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity4fidelity5style5cross page consistency4

Same girl: brown skin, two hair puffs, round red glasses, yellow raincoat, blue-and-white striped shirt, and green rubber boots all present. Minor drift: the raincoat appears mostly closed rather than clearly worn open as in the reference, and proportions are slightly more compact.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, children's book illustration style, warm colors, simple shapes
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity4fidelity5style5cross page consistency4

Same girl: brown skin, two curly black puffs, round red glasses, yellow raincoat open over blue/white striped shirt. Green boots are visible but partially obscured by the kneeling pose; still present. Minor drift only due to pose hiding some details.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, children's book illustration style, warm colors, simple shapes
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity4fidelity5style5cross page consistency4

Same girl: brown skin, two hair puffs, round red glasses, yellow coat, green rubber boots all present. The striped shirt is barely visible (coat is mostly zipped/closed rather than open showing stripes), which is a minor drift from the reference where the coat is open over a clearly visible striped shirt.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, children's book illustration style, warm colors, simple shapes
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity5fidelity4style5cross page consistency4

All reference-visible anchors are preserved: brown skin, black hair in two puffs, round red glasses, open bright yellow raincoat over blue-and-white striped shirt, blue shorts, and green rubber boots. Facial features and proportions match closely.

promptMaya, a young girl with brown skin and curly black hair in two puffs, round red glasses, a bright yellow raincoat worn open over a striped shirt, and green rubber boots, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, children's book illustration style, warm colors, simple shapes

Pepper

raccoonnotched left ear, star chest patch, green backpack
Pepper reference
referenceminted once · medium · /generationsfrozen canonical

Minted once, then reused for every scene, every model, every future run. Continuity is judged against this image.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, standing in a relaxed friendly pose, plain soft cream background, full body visible, children's book illustration style, warm colors, simple shapes

gpt-image-2

cross-page consistency4
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + referenceref gap
continuity4fidelity5style4cross page consistency4

Same gray raccoon with black mask, ringed tail, white star chest patch, and green backpack all preserved. Ears appear intact in both images (no notch visible in reference either). Proportions are slightly more chibi/large-headed in the new illustration, a minor drift.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, children's book illustration style, warm colors, simple shapes
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + referenceref gap
continuity4fidelity5style5cross page consistency4

Same gray raccoon with black mask, ringed tail, white star chest patch, and green backpack. Ears look intact in both images (no visible notch in either), so that's consistent with reference. Proportions slightly more top-heavy with larger head in the new illustration, but identity anchors visible in the reference are all preserved.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, children's book illustration style, warm colors, simple shapes
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + referenceref gap
continuity4fidelity5style5cross page consistency4

Same gray raccoon with black mask, ringed tail, white star patch on chest, and bright green backpack all present and matching. Ears both appear intact in reference (no visible notch), so that anchor isn't in reference. Minor drift: proportions are slightly chubbier/rounder in scene image, but clearly the same character.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, children's book illustration style, warm colors, simple shapes
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + referenceref gap
continuity4fidelity5style4cross page consistency4

Same gray raccoon with black mask, white star chest patch, ringed tail, and green backpack all present. Proportions are slightly more chibi/large-headed than the reference, but identity clearly holds. Note: the reference itself does not show a notched left ear.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, children's book illustration style, warm colors, simple shapes

nano-banana

cross-page consistency4
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + referenceref gap
continuity3fidelity5style4cross page consistency4

Same gray raccoon species with black mask, ringed tail, and green backpack preserved. However, the distinctive white star-shaped chest patch visible in the reference is missing/obscured in the new illustration. Ear notch is not clearly visible in either but is a sheet anchor. Overall recognizably the same character but a key reference-visible anchor (star patch) is lost.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, children's book illustration style, warm colors, simple shapes
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + referenceref gap
continuity4fidelity5style5cross page consistency4

Same gray raccoon with black mask, ringed tail, and green backpack. The white star chest patch is faintly visible. Ears both appear intact in both images--the reference does not clearly show a notched left ear, so that anchor is a reference gap, not a continuity break. Proportions and coloring match well.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, children's book illustration style, warm colors, simple shapes
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + referenceref gap
continuity4fidelity5style5cross page consistency4

Same gray raccoon with black mask, ringed tail, white star patch on chest, and green backpack all present and consistent. Proportions and coloring match. Minor drift: the mouth is open in excitement (expression change, acceptable) and the ear notch is not clearly visible, but the notch is also not clearly shown in the reference.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, children's book illustration style, warm colors, simple shapes
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + referenceref gap
continuity4fidelity4style4cross page consistency4

Same gray raccoon with black mask, white star patch on chest, ringed tail, and bright green backpack all present. Ears appear intact without a visible notch, but the reference itself didn't clearly show a notch either. Slight style/proportion drift but unmistakably Pepper.

promptPepper, a small gray raccoon with a notch missing from the left ear, a white star-shaped patch of fur on the chest, a black mask and ringed tail, wearing a bright green backpack, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, children's book illustration style, warm colors, simple shapes

Bolt

robotone big eye, star antenna, tank treads, red belly hatch
Bolt reference
referenceminted once · medium · /generationsfrozen canonical

Minted once, then reused for every scene, every model, every future run. Continuity is judged against this image.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, standing in a relaxed friendly pose, plain soft cream background, full body visible, children's book illustration style, warm colors, simple shapes

gpt-image-2

cross-page consistency4
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity4fidelity5style5cross page consistency4

Same teal round robot with one large eye, star-topped antenna, and tank treads clearly preserved. The red belly hatch is not visible (obscured by the book, plausibly hidden rather than removed), which is a minor drift but reference-visible anchor isn't confirmed present.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, children's book illustration style, warm colors, simple shapes
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity5fidelity5style5cross page consistency4

Teal round body, single large friendly eye, yellow-star antenna, tank treads, and red belly hatch all faithfully reproduced from the reference.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, children's book illustration style, warm colors, simple shapes
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity5fidelity5style5cross page consistency4

All reference-visible anchors are preserved: teal round body, one large friendly eye, antenna with yellow star, red hatch door on belly, tank treads, striped arms, and same cheerful smile.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, children's book illustration style, warm colors, simple shapes
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity5fidelity5style4cross page consistency4

Teal round robot body, single large friendly eye, yellow star antenna, red belly hatch, and tank treads are all preserved from the reference. Proportions and color palette match well.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, children's book illustration style, warm colors, simple shapes

nano-banana

cross-page consistency4
library
librarynano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity3fidelity5style4cross page consistency4

Same teal robot with one round eye, star-tipped antenna, and tank treads are preserved. However, the red hatch door on the belly is missing in the new illustration, which is a distinguishing anchor visible in the reference. Proportions and color are otherwise consistent.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, sitting in a cozy library reading nook, holding open a big picture book, an orange cat curled up nearby, children's book illustration style, warm colors, simple shapes
beach
beachnano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity4fidelity5style4cross page consistency4

Same teal round robot with one large eye, star-topped antenna, red belly hatch, and tank treads. Identity is unmistakable, though the treads have shifted to look more like wheels/single tread pair rather than the clearer dual tank treads seen in the reference -- a minor anchor drift.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, building a tall sandcastle on a sunny beach, two seagulls overhead, a red bucket and shovel in the sand, children's book illustration style, warm colors, simple shapes
sledding
sleddingnano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity5fidelity5style4cross page consistency4

All reference-visible anchors are preserved: teal round body, single large friendly eye, antenna with yellow star, tank treads, red belly hatch, and same segmented arms. Proportions match closely.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, riding a wooden sled down a snowy hill past pine trees, snowflakes falling, arms up with excitement, children's book illustration style, warm colors, simple shapes
garden
gardennano-banana (gemini-2.5-flash-image) · generateContent + reference
continuity4fidelity4style4cross page consistency4

Same teal robot with one large round eye, yellow star antenna, tank treads, and red belly hatch. However, the reference clearly shows distinct arms with segmented joints and hands, while the new illustration has arms that appear to hold watering cans but the second 'arm' looks duplicated/awkward - minor drift in limb structure.

promptBolt, a short round teal robot with one large friendly round eye, a single antenna topped with a small yellow star, tank treads instead of legs, and a little red hatch door on its belly, planting tomato seedlings in a community garden, holding a small watering can, raised wooden garden beds and a sunflower in the background, children's book illustration style, warm colors, simple shapes

Zuri

human childvoluminous curly hair, round black glasses, teal puffer vest, graphic tee, jeans
Zuri reference
referencecropped from the style-picker UIfrozen canonical

Not minted: a crop of the production style-picker's 3D Animation panel (~128px, upscaled) -- the realistic worst case for reference quality. Cropped mid-thigh, so footwear is not a judgeable anchor.

character sheetZuri, a young girl with brown skin and voluminous dark curly hair, round black-rimmed glasses, a bright teal puffer vest worn open over a cream graphic t-shirt, blue jeans, and sneakers

No default-style battery for this character yet -- her runs so far are the vibrant-fantasy arm on the styles page.

Access probes

the single-call checks run before each build phase
fox mint
fox mintgpt-image-1 · low · /generations

Text-only mint of the probe character.

fox, new scene
fox, new scenegpt-image-1 · low · /edits + reference

Same fox, new scene, first attempt -- the capability this ticket proves.

fox on gpt-image-2
fox on gpt-image-2gpt-image-2 · low · /generations

Access check only; full battery queued.

pepper on mini
pepper on minigpt-image-1-mini · low · /edits + reference

Budget tier takes references; anchors sloppier.

nano fox mint
nano fox mintnano-banana · generateContent

Key + endpoint confirmed (free tier has zero image allowance -- billing required).

nano fox, new scene
nano fox, new scenenano-banana · generateContent + reference

Reference path works -- unblocked the battery.

nano pepper
nano peppernano-banana · generateContent + reference

First cross-model look; the star chest patch went missing -- opus later confirmed the anchor class.

nano maya (child)
nano maya (child)nano-banana · generateContent + reference

The K-8 policy probe: a human child rendered from a reference with NO safety block.

Full written results

the complete documents behind this page -- method, every table, the reads, and the standing caveats

_Last updated: 2026-07-23_

Pass 2 -- character continuity from reference images: findings + recommendation

Scope of this pass: can a model take a student-built character's reference image and reproduce that character faithfully across new scenes? Candidates were Amazon Nova Canvas 1.0 and OpenAI GPT Image (gpt-image-1), per the pass-2 kickoff (deliberately overriding the ticket's earlier Nova Canvas rule-out to test its reference path; the Stability tier bake-off remains a separate, later track). Research pull date for all citations: 2026-07-22. Bedrock/AWS pricing pages are JS-rendered -- reconfirm any dollar figure in the console before it feeds a budget.

The short version

  1. Nova Canvas is out -- twice over. The test account is Legacy-locked out of it (verified by direct invoke), and even with access, its zero-shot reference modes are documented to carry style/palette/layout, not character identity. Its only documented identity mechanism is per-character fine-tuning, which Legacy status forbids starting ("you cannot create new fine-tuning jobs"), and the whole model EOLs 2026-09-30. There is no Nova Canvas story for this feature.
  2. GPT Image can do this. On a judged 3-character x 4-scene battery (opus-over-Bedrock judge, outside the generator's model family), reference-conditioned renders scored 4.42/5 average character continuity, 4.7/5 scene fidelity, ~5/5 style match, with per-character cross-page consistency of 4-5. The failures are legible and small: one hidden belly hatch, sub-anchor drift like star-patch crispness. Nothing looked like a different character.
  3. But do not ship on gpt-image-1. The OpenAI lineup moved: gpt-image-1 is deprecated (third-party sources cite retirement 2026-10-23 -- the same date the text-model baseline dies; reconfirm on OpenAI's deprecations page). The current flagship is gpt-image-2, which processes reference input at high fidelity automatically and prices slightly LOWER per token than gpt-image-1. A single access call confirmed this account can invoke gpt-image-2 today. The next slice should re-run this exact battery on it.
  4. The ceiling is elsewhere and worth knowing: Google's Gemini image models publish character consistency as a headline capability (Nano Banana Pro claims consistency for up to 5 people; ~20 reference images accepted on 2.5 Flash Image, extra references free), at ~$0.039/image for 2.5 Flash Image. Bedrock's only native answer, Nova 2 Omni, is still limited to Nova Forge customers in preview as of 2026-07. Neither was built this pass, per scope.

Access probe (Task 0)

Full detail in 2026-07-22_pass2-access-probe.md. Summary: Nova Canvas TEXT_IMAGE in us-east-1 returned the Legacy lockout ("...you have not been actively using the model in the last 30 days"); eu-west-1 was SCP-denied at the org level; gpt-image-1 passed on both /v1/images/generations and the reference-conditioned /v1/images/edits.

One documentation conflict worth keeping: the Bedrock model-lifecycle page publishes a 15-day inactivity clause for Legacy models, while the live denial names 30 days (lifecycle table, pull 2026-07-22). Cite the probe's 30-day error as the observed enforced value and the doc's 15-day as published policy; do not treat them as interchangeable.

Nova Canvas 1.0 -- the paper record (all pull 2026-07-22)

What it could have offered, per the Nova user guide (request structure, access page, overview):

  • IMAGE_VARIATION: 1-5 base64 reference images (images), similarityStrength 0.2-1.0, text 1-1024 chars. Documented to borrow "style, color palette, and subject" from references -- variation and style bias, not identity preservation.
  • TEXT_IMAGE + conditionImage: one reference, CANNY_EDGE or SEGMENTATION controlMode, controlStrength 0-1 (default 0.7) -- layout/composition/edges only.
  • COLOR_GUIDED_GENERATION: up to 10 hex colors + one optional subject/style referenceImage.
  • Subject consistency is a fine-tuning feature, verbatim: "Subject consistency is achieved by fine-tuning the model with reference images to preserve the chosen subject (for example, pet, shoe, or handbag) in generated images." AWS's own storyboard-consistency blog series concedes prompt+seed techniques "aren't perfect ... even images within the same scene show variations" and routes readers needing near-perfect consistency to fine-tuning + provisioned throughput (part 2; part 1's aws.amazon.com URL 404'd on fetch -- quotes corroborated via republications, reconfirm in console).
  • Legacy state forbids new fine-tuning jobs and new provisioned throughput (lifecycle), so the one documented identity path is closed even before the 2026-09-30 EOL. Per-student fine-tuning would also have been architecturally absurd for our use case.
  • For the record: pricing ~$0.04-$0.08/image by resolution/quality (secondary aggregation; JS-rendered primary), 98.1% benign pass rate on its published service-card evals (AI Service Card), regions us-east-1 / eu-west-1 / ap-northeast-1.

Verdict: closed permanently. Blocked in fact, incapable on paper, dead in September.

GPT Image -- judged results

Harness: the continuity-run script + lib/continuity-eval-set.mjs + lib/continuity-judge.mjs in the harness repo. Three student-character stand-ins (Maya: human child with hair puffs / red glasses / yellow raincoat / green boots; Pepper: raccoon with notched ear / star chest patch / green backpack; Bolt: teal robot with one eye / star antenna / treads / belly hatch), each frozen to one canonical reference, then rendered into 4 shared scenes (library, beach, sledding, garden) via /v1/images/edits with the reference attached. Judge: claude-opus-4-7 over Bedrock (different model family from the generator), scoring 1-5 on character continuity (against the reference image as ground truth), scene fidelity, and style match, plus a per-character cross-page consistency pass.

CharacterScenesContinuityFidelityStyleCross-page consistency
maya45, 5, 5, 55, 5, 5, 55, 5, 5, 55 (no worst page)
bolt43, 5, 5, 45, 5, 5, 45, 5, 5, 54
pepper44, 4, 4, 45, 5, 5, 45, 4, 5, 54
baseline overall124.42 avg4.75 avg4.83 avg4.33 avg
pepper, input_fidelity=high44, 4, 4, 45, 5, 5, 55, 5, 5, 54

What the misses actually were:

  • Bolt/library (the only 3): the red belly hatch went missing/hidden -- an occlusion-or-drop failure on a reference-visible anchor. The other robot dings were sub-anchor (eye slightly larger, arm banding).
  • Pepper's flat 4s: residual drift in star-patch crispness and head shape; every major anchor (mask, ringed tail, star patch, backpack) held in all 8 pepper renders across both arms.
  • input_fidelity=high bought nothing measurable on continuity at n=4 (4.00 vs 4.00 for the same character baseline), with scene fidelity ticking up to straight 5s. Docs say the param is gpt-image-1-only; gpt-image-2 does high-fidelity input automatically and the docs say to omit the parameter there (cookbook, image guide). One rule we already comply with and should keep encoded: when passing multiple references, only the FIRST image gets the extra-fidelity treatment -- the character reference must lead the array.
  • Methodology catch that matters for the product: the first pepper run scored 3s because the judge was docking the scenes for a sheet anchor (the ear notch) that the minted reference itself never showed. The scenes were faithful to the reference. The judge now treats the reference image as ground-truth identity and reports sheet-vs-reference gaps in a separate reference_gap field. Product implication: reference QC at character-creation time is its own quality gate -- whatever the student freezes is what continuity is measured against.

Run artifacts: the run archive in the harness repo (frozen references under references/, per-run images + results.jsonl + summary.md under runs/).

GPT Image -- capability + production facts (pull 2026-07-22)

  • Endpoints: references go to /v1/images/edits as multipart image[], up to 16 images, PNG/WebP/JPG, 50MB each; /generations is text-only (edit reference). Sizes 1024x1024 / 1536x1024 / 1024x1536 / auto; quality low/medium/high/auto; moderation auto|low.
  • Pricing (token-based): gpt-image-1 $5/1M text-in, $10/1M image-in, $40/1M image-out; practical per-image at 1024x1024 = $0.011 low / $0.042 medium / $0.167 high, portrait/landscape higher ($0.25 at high). A reference image bills as image-INPUT tokens on top of the output cost, more under high fidelity. gpt-image-2 is cheaper per token ($8/1M in, $30/1M out); gpt-image-1-mini is the budget tier ($2.50/1M in, $8/1M out) (pricing, corroborated by pricepertoken.com).
  • Deprecation: gpt-image-1 is deprecated; current lineup is gpt-image-2 (snapshot 2026-04-21), gpt-image-1.5, gpt-image-1-mini (model pages). The 2026-10-23 gpt-image-1 retirement date is third-party -- reconfirm. Access to gpt-image-2 from this account was confirmed by a live generation call 2026-07-23 (the run archive) -- an access check only; no eval was run on it this pass, per the two-model scope.
  • Moderation risk is real for us: blocks surface as HTTP 400 moderation_blocked, moderation: "low" reduces but does not remove strictness, and there are documented benign false positives on safe images of children (community thread, Azure-variant report). None of our 21 kid-scene calls this pass got blocked, but a benign-block battery (the ticket's experiment C, pointed at the OpenAI side) is the right next measurement.
  • Rate limits cap production: images/min by tier -- Tier 1: 5, Tier 2: 20, Tier 3: 50, Tier 4: 150, Tier 5: 250. A 6-page story is 6 renders; per-student real-time generation needs tier planning or queuing.
  • Third-party honesty check: developers report character inconsistency even with input_fidelity=high in some cases (community); no published subject-consistency benchmark exists for gpt-image-1. Our judged 4.42/5 is best-effort evidence on our own kind of content, not a guarantee.

The ceiling (research only, not built)

  • Gemini 2.5 Flash Image ("Nano Banana," gemini-2.5-flash-image) publishes character consistency as a headline capability, takes up to ~20 reference images (10MB each, extra references not billed), ~$0.039/image (Google blog). Its successor Nano Banana Pro (gemini-3-pro-image, preview, 2025-11-20) blends up to 14 images, claims consistency for up to 5 people, 2K/4K output at ~$0.13-$0.24/image, and holds a top LMArena image rank per secondary trackers (DeepMind page). This is the published leader for exactly our task and the benchmark any GPT Image result should eventually be compared against.
  • Nova 2 Omni remains restricted to Nova Forge customers via AWS account teams as of 2026-07 (what's new); it advertises character consistency but has no public model id, schema, or benchmark. Still the only Bedrock-native watch item, unchanged from the ticket.

Recommendation

Character continuity from a student reference is viable today on the OpenAI side, and only there among what we can invoke. Concretely:

  1. Treat pass 2 as proof of capability: reference-conditioned edits held identity at 4.42/5 with legible, minor failure modes on exactly the content we care about.
  2. Point the next slice at gpt-image-2: the same frozen references, byte-identical scene prompts, same judge -- plus a benign-block battery and a cost read (access already confirmed; lower per-token price and reference fidelity is automatic). Do not invest further in gpt-image-1 -- it is deprecated with a likely October retirement.
  3. Keep Bedrock in the picture only via the watch list (Nova 2 Omni GA) and the separate Stability tier bake-off; neither addresses reference-based identity today.
  4. Carry the two production risk items into any build ticket: moderation false positives on child content (measure, then decide on moderation: "low"), and images/min tier limits.
  5. Product-side: reference QC at character-creation time is a required quality gate -- continuity is measured against whatever the student freezes.

Blocked / unconfirmed / not done

  • Nova Canvas IMAGE_VARIATION was never live-tested (access blocked); its style-not-identity characterization is documentary.
  • gpt-image-1's 2026-10-23 retirement and gpt-image-2 per-image dollar costs are third-party -- reconfirm before budgeting.
  • Nova Canvas seed and numberOfImages exact ranges were truncated in the doc fetch; the 15-vs-30-day Legacy inactivity discrepancy is unresolved.
  • Not tested this pass: gpt-image-2 quality (access only), multi-character scenes, student-drawn (rough) reference art, cross-run reference reuse at scale, benign-block rates, Gemini anything.

_Last updated: 2026-07-22_

Pass 2 -- access probe results (Nova Canvas 1.0 + GPT Image)

Before building anything, one direct invoke against each candidate settled whether the test account can reach them at all. Probe script: the access-probe script in the harness repo; outputs under the run archive.

Verdict table

ProbeModel / pathResult
1aNova Canvas amazon.nova-canvas-v1:0, TEXT_IMAGE, us-east-1BLOCKED -- ResourceNotFoundException: Access denied. This Model is marked by provider as Legacy and you have not been actively using the model in the last 30 days. Please upgrade to an active model on Amazon Bedrock
1a-retrySame model, eu-west-1BLOCKED -- AccessDeniedException with an explicit deny from the org service control policy (non-US region; SCP p-pkbqi430)
1bNova Canvas IMAGE_VARIATION (the reference-image path)SKIPPED -- unreachable behind 1a
2agpt-image-1 /v1/images/generationsPASS -- 1024x1024 image returned in 36s
2bgpt-image-1 /v1/images/edits with a reference imagePASS -- returned the same cartoon fox (face, scarf, palette, style) in a new scene on the first attempt

What this means

  • The Nova Canvas arm is dead for us. The ticket's warning ("Legacy access rules may block the test account from even testing it") is now a verified fact, with the exact denial text captured above. Note the lockout window in the live error is 30 days of inactivity, not the 15 days our survey doc carried -- worth a one-line correction there. Even if access were granted, the model EOLs 2026-09-30, so nothing durable could be built on it anyway.
  • The Bedrock gap is real and structural. With Titan Image v2 past EOL, Nova Canvas Legacy-locked, and Nova 2 Omni restricted to Nova Forge customers in preview, Bedrock currently offers us no self-serve model with a reference-image path. Reference-conditioned character continuity has to come from outside Bedrock today; the realistic Bedrock-native options are Stability image-to-image (a weaker, style-level lever) or waiting on Nova 2 Omni GA.
  • GPT Image is fully reachable on the existing harness key, both plain generation and the reference-conditioned edits path, so pass 2 proceeds as a GPT Image evaluation with Nova Canvas documented as blocked.

Rerun

node the evaluation harness/the access-probe script

AWS creds come from the credentials file (fresh test account SSO), the OpenAI key from .env.harness. NOVA_CANVAS_REGION overrides the Nova region if anyone wants to re-verify the denial.