Summary for Anime & Cartoon Style

GPT Image 2.6 Sunburst leads at 1362 Elo, followed by Seedream 5.5 at 1337 Elo and Seedream 5.0 Pro at 1331 Elo. These form a closely spaced leading group: first to third is only 31 points, so the ordering should not be mistaken for decisive separation.

Leading models by category Elo

Model Category Elo ↓ Mean image score* Images returned
GPT Image 2.6 Sunburst 1362 Elo 8.57/10 7/10
Seedream 5.5 1337 Elo 8.00/10 8/10
Seedream 5.0 Pro 1331 Elo 8.14/10 7/10
GPT Image 2 1294 Elo 8.29/10 7/10
Nano Banana 2.1 1294 Elo 8.00/10 9/10
GPT Image 2.5 Sunburst 1287 Elo 8.22/10 9/10
MAI-Image-2.6 1267 Elo 8.17/10 6/10
Muse Image 1254 Elo 7.40/10 10/10

Successful images only. These averages explain output quality; they do not determine the ranking.

Quick conclusions

  • Highest-rated starting point: GPT Image 2.6 Sunburst, particularly for expressive anime characters, readable action, and detailed costume design. Allow for non-delivery on some briefs.
  • Strong illustrated storytelling: Seedream 5.5, whose standout examples include the animal adventure and sleeping chibi dragon. Its results are less consistent across everyday scenes and retro spacecraft.
  • Highest-rated model with complete delivery in this sample: Muse Image. It returned all ten images and offers an attractive alternative when uninterrupted coverage matters.
  • Fast iteration: Nano Banana 2 Lite combines 1190 Elo with a roughly 3.1-second median successful generation time and nine completed images.
  • Strong balance of rating, speed, and coverage: Nano Banana 2.1, at 1294 Elo, returned nine images with a roughly 22.1-second median.

The central lesson is simple: clear anatomy, coherent object interaction, faithful styling, and understandable storytelling matter more than sheer decorative density.

General Analysis & Useful Insights

1. The leaders succeed in different ways

GPT Image 2.6 Sunburst: strong character design and explicit action. Its successful outputs average 10/10 for prompt adherence. The evaluation rewards details that reinforce the subject rather than merely covering surfaces. In the samurai portrait, patterned robes, ribbons, and armor create a distinctive key-art identity, although the sword-side hand remains crowded. In the retro space battle, visible gun emissions, opposing ship families, and distributed impacts make the engagement unusually easy to follow. The rabbit trick similarly combines the rabbit’s relaxed confidence with a readable slapstick consequence.

Seedream 5.5: composition and environmental storytelling. The samurai portrait integrates the turning figure, railing, canal, and castle into clear spatial layers. The cat-and-dog adventure makes the travelers feel situated within a larger journey, while the sleeping dragon uses books, a pocket watch, and an embroidered cushion to create a memorable setting. This is not uniform dominance: the evaluation finds weaker hand-and-object construction in the ramen scene and less convincing period styling in the space battle.

Seedream 5.0 Pro: material richness and inhabited fantasy worlds. Its successful outputs average 9/10 for artistic merit. Domestic touches, terraces, and surrounding geography strengthen the floating castle. Carefully differentiated scales, fabric, jewels, and reflected light distinguish the sleeping dragon. However, the magical-girl scene gives the environment more authority than the requested character design, and the rabbit scene shifts toward plush 3D animation rather than classic cartoon styling.

Practical implication: the top three are close in Elo, but not interchangeable. Character action, illustrated storytelling, and dimensional fantasy rendering should guide which one you try first.

2. Close contenders offer meaningful practical alternatives

GPT Image 2 and Nano Banana 2.1 are tied at 1294 Elo. GPT Image 2.5 Sunburst, at 1287 Elo, belongs in the same close comparison group. Their different mean image scores should not break that Elo tie.

A second useful band spans just 28 points: Qwen-Image-3.0-Pro at 1201 Elo, Grok Imagine 2.0 at 1195 Elo, Nano Banana 2 Lite at 1190 Elo, FLUX 3 Image at 1181 Elo, and Nano Banana 2 at 1173 Elo. Treat these as close and choose by requirements:

3. What separates convincing images from decorative ones?

Readable interaction is a major quality bottleneck

The evaluation repeatedly penalizes errors precisely where the requested action occurs:

Interaction Reported weakness Why it matters
Magical girl casting a spell An extra digit on the enlarged casting glove The anatomical defect dominates the spellcasting focal point.
Cat and dog adventuring An additional cat paw Crisp scenery cannot repair a primary character’s construction.
Character eating ramen A reversed soup spoon Strong atmosphere is undermined by incorrect everyday behavior.
Rabbit tricking a hunter The cut rope does not clearly release the anvil The joke depends on an understandable mechanism.

This explains an important result: GPT Image 2.6 Flare averages 9.43/10 adherence, but only 5.86/10 coherence, and ranks at 1052 Elo. It returned the same seven prompts as GPT Image 2.6 Sunburst, so their large rating difference cannot be attributed solely to different completion coverage. Its strong rabbit-trap illustration is a useful exception, not evidence that the wider anatomical weaknesses disappear.

Style fidelity is not the same as rendering polish

A glossy 3D interpretation can be appropriate for the chibi dragon prompt, which does not explicitly demand flat 2D rendering. The same finish conflicts with the colorful 2D animal adventure or classic cartoon princess.

Similarly, hatching, distressed grain, and cinematic lighting are not substitutes for the explicit dot-based shading in the halftone superhero prompt. The evaluation distinguishes these techniques rather than treating all vintage-looking texture as equivalent.

Narrative verbs need visible evidence

For the rabbit trick prompt, merely showing a smiling rabbit beside an armed hunter is insufficient. Stronger examples establish misdirection, a consequence, or both. The pitfall scene gives the viewer an immediate cause-and-effect sequence. The bent-shotgun scene uses deliberate cartoon impossibility as the joke, rather than presenting it as an accidental construction error.

The same principle applies to the sleeping dragon: an open or winking eye can materially weaken adherence, however polished the scales and treasure may be.

More texture can reduce clarity

Midjourney V6.1, at 793 Elo, averages 7.8/10 for artistic merit but 5.9/10 for coherence. Midjourney v7, at 735 Elo, shows a similar gap: 7.3/10 artistry versus 5.5/10 coherence. Their imagery can be atmospheric and inventive, but crowded anatomy, unclear attachments, and subject drift recur in the supplied evaluation.

For complex machinery, the successful distinction is not simply more panel lines. In the retro space battle, stronger scenes separate engines from weapons and connect shots with identifiable impacts. In the floating castle, readable terraces, supports, and planted areas matter more than an indiscriminate mass of cogs and cables.

4. Delivery and speed change the practical shortlist

31 of the 34 no-image outcomes occur across the floating castle, classic princess, superhero, and rabbit trick briefs. The error records indicate 26 moderation-related failures and eight technical or validation failures. These records do not establish a legal conclusion or a universal policy rule about the prompts.

For example, Seedream 5.0 Pro has three recorded generation failures rather than three explicit moderation refusals. The single failed outputs from Nano Banana 2 Lite and Nano Banana 2.1 are validation failures.

Model Category Elo Successful images Median successful time
GPT Image 2.6 Sunburst 1362 Elo 7/10 154.0 s
Seedream 5.5 1337 Elo 8/10 24.4 s
Seedream 5.0 Pro 1331 Elo 7/10 114.9 s
Nano Banana 2.1 1294 Elo 9/10 22.1 s
Muse Image 1254 Elo 10/10 12.7 s
Nano Banana 2 Lite 1190 Elo 9/10 3.1 s
Nano Banana 2 1173 Elo 10/10 16.5 s

Medians conceal occasional long waits. Seedream 5.0 Pro, for example, records approximately 938 seconds for the magical girl and 748 seconds for the rabbit trick. Recorded durations are operational observations, not controlled hardware benchmarks or evidence of monetary cost.

Best Model Analysis by Use Case

Category Elo remains the model ranking. No separate portrait, storybook, comic, or retro-anime Elo is supplied. The shortlists below combine that ranking with observed style fit and delivery. Individual examples are evidence to inspect—not a new ranking based on their 1–10 scores.

Anime characters and expressive costume design

Samurai beneath cherry blossoms

Start with GPT Image 2.6 Sunburst, 1362 Elo, and Seedream 5.5, 1337 Elo. These close leaders offer different emphases: Sunburst’s portrait has especially distinctive ribbons, fabric motifs, and key-art styling; Seedream’s portrait emphasizes coherent anatomy, layered armor, and spatial integration.

For this brief, inspect hands, sword guards, scabbards, and armor attachment points before judging the finish. Beautiful blossoms and facial drawing can conceal equipment problems.

Magical girl casting a spell

The Elo-led shortlist is GPT Image 2.6 Sunburst, Seedream 5.5, and GPT Image 2.5 Sunburst, at 1362 Elo, 1337 Elo, and 1287 Elo, respectively.

Compare Seedream’s buoyant celestial design with the 2.5 Sunburst casting scene. The latter provides particularly readable foreshortening and wand handling. For rapid concept exploration, Nano Banana 2 Lite, 1190 Elo, supplies a strong full-body rooftop example.

Prioritize a traceable casting gesture and separated costume layers. More spell circles do not necessarily improve the result.

Colorful cartoons and visual storytelling

Cat and dog on a colorful 2D adventure

Consider GPT Image 2.6 Sunburst, Seedream 5.5, and Nano Banana 2.1, in descending category Elo.

Seedream’s sunset journey demonstrates controlled environmental storytelling without relying on excessive action. Nano Banana 2.1’s forest scene has a cohesive storybook identity and expressive gestures. Nano Banana 2 Lite’s bridge crossing is worth inspecting when iteration speed matters.

Look for distinct animal identities, readable limb counts, purposeful equipment, and a route or destination. A backpack alone can leave the adventure feeling generic.

Classic 2D princess singing with animals

Muse Image, 1254 Elo, is the highest-rated category model that returned an image for every prompt, including this one. Its princess illustration makes singing explicit through expression, gesture, and musical notation.

FLUX 3 Image, 1181 Elo, provides an especially relevant traditional-animation example, with coherent animal placement and a lyrical musical staff. For period-style reference, also inspect Nano Banana Pro, 1104 Elo, and its vintage forest performance.

Choose based on the flat-character/painted-background relationship and whether the scene reads as singing rather than conversation. Several higher-rated category models returned no image for this particular brief.

Rabbit tricking a hunter

Start with GPT Image 2.6 Sunburst, 1362 Elo, then consider Nano Banana 2.1, 1294 Elo. The hunting-season reversal shows a clear slapstick consequence, while the bent-shotgun gag makes the sabotage immediately visible.

For a concise alternative, Grok Imagine, 1033 Elo, offers a clearly staged pitfall. This is a specialist example from a lower-rated category model, not grounds to promote it above the category leaders.

The key requirement is readable deception. Signs, dialogue, and exaggerated expressions should explain or support the gag—not replace it.

Backgrounds, atmosphere, and everyday life

Floating steampunk castle with watercolor textures

For models that completed this brief, inspect Seedream 5.0 Pro, 1331 Elo, Nano Banana 2.1, 1294 Elo, and GPT Image 2.5 Sunburst, 1287 Elo.

Seedream’s castle combines inhabited terraces and small domestic details with an expansive surrounding world. The 2.5 Sunburst scene emphasizes monumental scale and cinematic staging, but has less explicit watercolor texture. Qwen-Image-3.0-Pro, 1201 Elo, supplies another strong castle-and-nature example and completed all ten category prompts.

Decide whether you value visible watercolor handling, cinematic scale, or intricate architecture most. They are related qualities, but the evaluation does not treat them as interchangeable.

Character eating ramen at a small shop counter

The high-Elo starting points include GPT Image 2.6 Sunburst, 1362 Elo, and GPT Image 2.5 Sunburst, 1287 Elo. MAI-Image-2.6, 1267 Elo, is another close contender to inspect for this specific setting.

The 2.5 Sunburst meal has a particularly clear relationship between hands, utensils, noodles, and mouth. MAI’s rainy-window scene adds restrained emotion and carefully differentiated everyday materials. Its excellent individual result should be considered alongside the model’s six-of-ten category completion, not in isolation.

For slice-of-life work, natural behavior and selective detail often communicate more than a heavily decorated kitchen.

Comics, retro action, and creature illustration

Superhero flying over a nighttime city with halftones

GPT Image 2 and Nano Banana 2.1 share 1294 Elo; GPT Image 2.5 Sunburst follows closely at 1287 Elo. These are sensible starting points among models that completed the brief.

Nano Banana 2.1’s panel demonstrates integrated dot shading, clear motion, and organized architectural detail. FLUX 3 Image, 1181 Elo, offers another strong heroine-and-city example.

Demand actual halftone shading, not only grain or hatching. Also inspect the pelvis, knees, and trailing boots: lower-body compression is a recurring weakness across otherwise polished superhero images.

Cute chibi dragon sleeping on gold

The leading completed candidates are Seedream 5.5, 1337 Elo, and Seedream 5.0 Pro, 1331 Elo—only six points apart.

Compare Seedream 5.5’s intimate study setting with Seedream 5.0 Pro’s pearlescent cave scene. The former emphasizes narrative props and varied materials; the latter emphasizes dimensional softness and reflected light. For a lighter illustrated approach, inspect Nano Banana 2.1’s coin-filled nest.

Here, a toy-like or dimensional treatment can fit the brief. Closed eyes, a supported resting pose, compact proportions, and readable treasure matter more than enforcing flat 2D rendering.

90s anime space battle with giant starships

Start with GPT Image 2.6 Sunburst, 1362 Elo. Other high-rated completed candidates include GPT Image 2, 1294 Elo, and GPT Image 2.5 Sunburst, 1287 Elo.

The 2.6 Sunburst battle makes firing directions and opposing formations unusually explicit. FLUX 3 Image, 1181 Elo, supplies a strong mechanically coherent retro fleet. A lower-rated specialist example is Flux 2 Pro, 1012 Elo, whose capital-ship engagement demonstrates convincing scale and period styling.

Check capital-ship scale, weapon origins, impacts, engine direction, and restrained cel shading. A fighter formation crossed by bright streaks is not necessarily a convincing giant-starship battle.

Practical selection guide

  • Maximum category rating: GPT Image 2.6 Sunburst, with a fallback for non-delivery.
  • High-rated illustrated storytelling with shorter recorded waits: Seedream 5.5.
  • Highest-rated complete coverage in this sample: Muse Image.
  • Very fast concept iteration: Nano Banana 2 Lite.
  • Fast iteration with all ten prompts completed: Nano Banana 2, whose 1173 Elo is close to Lite’s 1190 Elo.

Before selecting a final image, inspect the requested action first, then anatomy and object contact, then style fidelity, and only afterward decorative richness.