Head to head: Wan vs Krea 3 Edit
Wan vs Krea 3 Edit
By Ryan Merket · Published
A close image-model matchup across emotional storytelling, botanical detail, strict visual constraints, and exact counts. The results hinge on whether polish and prompt-specific details matter more than a few standout scene-building wins.
Krea 3 Edit takes the saved result, with an aggregate score of **65.1** to Wan’s **61.4**. The task ledger is **5 wins for Krea, 2 for Wan, and 1 tie**—a clear result on this sample, not a claim about either model beyond it. Krea’s edge comes from repeatedly nailing the requested read of an image: it makes the greenhouse keeper’s relief feel earned, gives the scabiosa the more convincing form and dew detail, and handles the seven-cup count with a cleaner presentation. It also better follows the restricted-palette brief. On the negation task, though, both images show foliage outside the window, so Krea’s win is about composition rather than fully satisfying the constraint. Wan has two meaningful wins. Its woven-apron scene includes the requested objects with crisper, more readable detail, while its oxide-palette allotment more completely delivers the poster brief, including the puddled stepping stones and allotment text. The perspective task is a tie: Krea’s aisle recedes convincingly, but Wan’s warmer lighting also serves the scene. **Verdict: Krea 3 Edit wins this saved task sample, with a clear result.** Wan is the better pick in these results for detailed, fully specified scenes; Krea is more consistent across emotional direction, botanical specificity, and clean constraint-following. This assessment applies only to the saved task sample.
Relieved greenhouse keeper
A cinematic portrait of a greenhouse keeper in mud-flecked overalls holding a cracked clay pot against his chest, his face unmistakably showing relieved disbelief after spotting the first bloom on a long-failing night-blooming cereus, with damp eyes, loosened jaw, and a half-laugh breaking through tension; three-quarter view, shallow-bokeh portrait lens, warm sodium grow-lights mixed with cool moonlight through glass panes, realistic painterly style, 16:9.
OpenAI: GPT-6 Luna: Model B more clearly conveys damp-eyed relieved disbelief and the half-laugh, with dramatic warm grow-light and cool moonlight. Model A has a slightly cleaner, wider greenhouse composition and more natural-looking hands, but its expression reads as straightforward happiness rather than release after tension. (Second judge pass, order swapped — scores are the average of both: Model B better captures the damp-eyed, tension-breaking relief and delivers the requested cinematic warm-and-cool greenhou | Anthropic: Claude Fable 5.1: Model B nails the emotional brief—damp eyes wit…
Woven apron and furrows
An ultra-detailed realism painting of a gardener’s workbench with a folded indigo linen apron, a coil of jute twine, a tarnished brass misting nozzle, a clump of freshly pulled rainbow carrots with hair-fine roots, and a tray of mossy seedling plugs, every surface rendered crisply so the fabric weave, twisted fibers, metal pitting, soil granules, leaf veins, and carrot skin ridges remain distinct without smudging; side-lit by a narrow greenhouse window, intimate close composition, 16:9.
OpenAI: GPT-6 Luna: Model A includes every requested object, with clear greenhouse side-lighting and crisp, readable details throughout. Model B has a compelling intimate, painterly look, but its heavy grain obscures some fine detail and the greenhouse window is not visible. (Second judge pass, order swapped — scores are the average of both: Model B has especially vivid rainbow carrots and richly emphasized surface texture, but its heavy grain and darker lighting make the scene feel less crisp and greenhouse-lit. M | Anthropic: Claude Fable 5.1: Model A delivers every requested element cohere…
Negation
A cozy reading nook with an armchair, a stack of books, and a mug of tea by a window — with absolutely NO plants, NO lamps, and NO artwork or picture frames anywhere in the frame. Warm afternoon light, 16:9.
OpenAI: GPT-6 Luna: Both images include visible foliage outside the window, violating the no-plants constraint, though neither shows indoor plants, lamps, or artwork. Model B has a more balanced, inviting composition, while Model A places the books and tea more naturally for the prompt. (Second judge pass, order swapped — scores are the average of both: Model B has a particularly warm, inviting composition, with the books and mug together on a table beside the window; Model A also fits the cozy reading-nook idea, b | Anthropic: Claude Fable 5.1: Both models satisfy the core negation (no plant…
Oxide palette allotment
A vintage travel poster illustration of the Brindle Fen community allotments after rain, showing a row of bean poles, a rusted wheelbarrow, puddled stepping stones, and broad rhubarb leaves under a low sky, rendered using only the named palette terracotta, sage green, cream, charcoal, and muted teal with no other colors anywhere in the image; bold screen-printed shapes, clean poster composition, soft overcast lighting, 16:9.
OpenAI: GPT-6 Luna: Model A more fully captures the vintage poster brief, with clearly visible puddled stepping stones, bean poles, a rusted wheelbarrow, broad rhubarb leaves, and correctly rendered allotment text. Model B has an appealing expansive composition and the requested palette, but its stepping stones are missing and the scene reads less clearly as a poster. (Second judge pass, order swapped — scores are the average of both: Model A more completely includes the puddled stepping stones and presents the sce | Anthropic: Claude Fable 5.1: Model A delivers a true poster layout with clea…
Macro dew anther
A macro close-up photograph of a single 'Blue Tango' scabiosa bloom at dawn, framed so the curled violet-blue petals spill diagonally across the image and the central anthers are razor-sharp while the background falls into creamy lens blur, with three dew droplets clinging to one filament and reflecting a tiny greenhouse roof; natural backlight from low sunrise, ultra-shallow depth of field, realistic botanical photography, 16:9.
OpenAI: GPT-6 Luna: Model B more clearly resembles a scabiosa and includes three droplets on the central filaments with greenhouse-like reflections, while Model A has only two prominent droplets and a less convincing botanical structure. Model A has appealing diagonal petals, but Model B better balances prompt-specific detail, shallow depth of field, and natural backlighting. (Second judge pass, order swapped — scores are the average of both: Model B more closely resembles a scabiosa and includes three dew droplets | Anthropic: Claude Fable 5.1: Model B depicts a convincing scabiosa with its…
Restricted palette
A flat-vector illustration of a mountain campsite at dusk using ONLY four colors — cream, burnt orange, teal, and charcoal — with no gradients and no stray off-palette colors, balanced composition, 16:9.
OpenAI: GPT-6 Luna: Model B better delivers the clean, flat-vector dusk scene with a restrained palette and crisp shapes, though its campsite is less detailed. Model A includes more recognizable camping gear, but its textured, subtly varied fills are less faithful to the no-gradients and strict-palette constraints. (Second judge pass, order swapped — scores are the average of both: Model B presents a clear, balanced mountain campsite in a clean flat-vector style and stays closer to the four-color restriction. Model | Anthropic: Claude Fable 5.1: Model A holds strictly to exactly four flat col…
Perspective & scale
A photorealistic one-point-perspective shot down a long empty library aisle, shelves receding to a single vanishing point, books getting consistently smaller with distance and nothing warped or wrongly sized, warm overhead light, 16:9.
OpenAI: GPT-6 Luna: Model B has a particularly clear, symmetrical one-point perspective, with the shelves and overhead lights converging convincingly toward the distant window. Model A is brighter and warmly lit, but Model B feels more immersive and its receding scale is slightly more coherent. (Second judge pass, order swapped — scores are the average of both: Model B creates a stronger sense of a long, empty aisle, with shelves and overhead lights converging cleanly toward a clear vanishing point. Model A is warm | Anthropic: Claude Fable 5.1: Model A delivers the bright warm overhead light…
Exact counting
A clean overhead flat lay on white marble of EXACTLY seven identical espresso cups arranged in a single straight row, each clearly separate and individually countable, even soft lighting, 16:9.
OpenAI: GPT-6 Luna: Both images show exactly seven clearly countable espresso cups in a straight row on white marble. Model B has softer, more even lighting and a cleaner, more centered presentation, while Model A’s marble veining and stronger shadows make the scene slightly busier. (Second judge pass, order swapped — scores are the average of both: Model B presents exactly seven clearly separated, identical cups in a particularly clean, straight row with soft, even lighting. Model A also follows the count and layo | Anthropic: Claude Fable 5.1: Both Model A and Model B render exactly seven i…
Matchup powered by OpenRouter.