Head to head: Wan vs Krea 3 Edit
Wan vs Krea 3 Edit
By Ryan Merket · Published
Eight image-generation challenges put prompt adherence, visual polish and text handling under the microscope. The results split between detailed scene execution and a few sharply defined strengths.
Wan’s edge came from getting more of the brief onto the canvas. It won five tasks, with its strongest showings in the apricot-hour bus stop and the palette-constrained alley: the requested actions, props and scene details were more consistently present. It also handled the balloon vendor’s specified hand roles better, despite misspelling the arcade sign. Krea 3 Edit took three wins, each for a clear reason. Its “NIGHT MARKET” poster nailed the exact three-line layout; its chrome reflections were more legible and dramatically lit; and its rain-soaked tram repair scene made the midnight atmosphere convincing. Those are meaningful wins, but they didn’t add up to broader consistency across the set. The aggregate scores are virtually tied, with Krea narrowly ahead, 60.2 to 59.9. That keeps this from being a runaway: Wan’s limited confidence result is a lean, not a license to call it decisively better. Both models had visible misses, from awkward composition and palette drift to grain and unclear details. **Verdict: Wan wins on task coverage and more reliable prompt execution, but only narrowly. Choose Krea 3 Edit when precise typography or reflective surfaces are the priority; for varied prompts with many specific constraints, Wan is the safer pick.**
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: Model A has the requested cozy chair, books, steaming tea, and warm window light, but the oversized chair and cropped book stack make the composition feel crowded. Model B presents the nook more clearly and with stronger balance, while both keep lamps and artwork out of the interior; greenery is visible outside the windows in both images. (Second judge pass, order swapped — scores are the average of both: Model B better respects the strict exclusions: the outdoor foliage is less prominent, while | Anthropic: Claude Fable 5.1: Both outputs honor the negation constraints (no…
Apricot Bus Stop Hour
A cinematic street scene at golden hour: a nearly empty bus stop on Larkspur Avenue with a glass shelter, one elderly woman in a mustard raincoat reading route map 9C, and long apricot-colored sunlight slicing between low buildings to illuminate suspended dust, warm metal bench edges, and the tops of parked scooters; mood should feel hushed and fleeting, with deep lengthening shadows, subtle lens glow, realistic painterly style, wide composition, 16:9.
OpenAI: GPT-6 Luna: Model A more fully captures the requested golden-hour scene, with the woman actively reading a route map, strong lengthening shadows, warm bench highlights, and visible suspended dust. Model B has a clearer Larkspur Avenue sign and a polished, balanced composition, but its light feels more diffuse and its woman is studying the shelter map rather than reading one herself. (Second judge pass, order swapped — scores are the average of both: Model A more clearly shows the elderly woman reading the r | Anthropic: Claude Fable 5.1: Model A delivers the cinematic apricot golden-h…
Rain-Slick Tram Repairs
A hyperreal editorial-style street photograph of a midnight tram repair crew in the Old Kestrel district, centered on a kneeling mechanic replacing a copper junction box beside track marker 47B; render razor-sharp fine texture in the oil-stained canvas coveralls, frayed reflective tape, beaded rain on brushed steel tools, braided cable sheathing, pitted asphalt, moss in the rail seams, and the tiny stamped lettering on a chipped teal service cart, with no smudged detail; cool sodium-vapor street lighting, shallow puddle reflections, low three-quarter angle, 16:9.
OpenAI: GPT-6 Luna: Model A more clearly centers the kneeling repair worker and sharply renders the copper junction box marked 47B, wet tools, cables, and teal cart, though the cart lettering misspells the district name. Model B conveys the midnight rain and mossy track seams well, but its image is grainier and the repair scene is less legible. (Second judge pass, order swapped — scores are the average of both: Model A more convincingly captures the rainy midnight repair scene, with a clear 47B box, braided cables, | Anthropic: Claude Fable 5.1: Model B delivers a convincing midnight editoria…
Balloon Vendor Portrait
A realistic urban portrait of a young balloon vendor standing outside the shuttered Mirren Arcade, shown full torso and clearly facing camera, holding exactly seven helium strings in the left hand and counting coins with the right hand so both hands are fully visible with natural finger anatomy; accurate shoulders, elbows, wrists, and body proportions, relaxed stance, wind-tugged jacket, candid documentary style, soft overcast morning light, background of faded posters and a red parking meter, medium shot, 16:9.
OpenAI: GPT-6 Luna: Model A better matches the requested hand roles, candid overcast portrait, faded posters, and red parking meter, though its arcade sign is misspelled. Model B renders the arcade name correctly but appears to swap the hands holding the strings and counting coins, and its lighting is less consistent with the soft overcast request. (Second judge pass, order swapped — scores are the average of both: Model A more clearly shows the requested hand roles, full torso, coins, posters, shutter, and red par | Anthropic: Claude Fable 5.1: Model B renders the 'Mirren Arcade' sign correc…
Reflections & glass
A photorealistic close-up of a chrome teapot on a polished dark marble counter next to a glass of water; the window and a red apple on the counter must be correctly reflected in BOTH the chrome and the water's surface, dramatic side light, 16:9.
OpenAI: GPT-6 Luna: Model B more clearly shows the window and a red apple reflected in the chrome, with stronger dramatic side light and a polished, photorealistic composition. Neither image clearly shows both required subjects reflected in the water’s surface, though Model A also has appealing counter reflections. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows the window and red apple reflected in the chrome, with dramatic side lighting and a polished dark counter. N | Anthropic: Claude Fable 5.1: Model B clearly shows the red apple and window…
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 A hews more closely to the requested cream, burnt-orange, teal, and charcoal palette, but its textured background, crowded campsite, and awkward horizontal bands weaken the flat-vector finish. Model B has a clearer, more polished composition and strongly evokes a dusk campsite, though its bright orange and multiple teal shades are less faithful to the strict four-color palette. (Second judge pass, order swapped — scores are the average of both: Model A has a clearer, more balanced campsite | Anthropic: Claude Fable 5.1: Model A keeps closer to the four-color brief wi…
Legible multi-line text
A minimalist event poster with three lines of crisp, correctly-spelled text stacked and centered: 'NIGHT MARKET' large on top, 'Fridays · 6–11pm' in the middle, 'Riverside Pier 4' at the bottom, on a deep navy background, clean sans-serif, subtle grain.
OpenAI: GPT-6 Luna: Model B follows the requested three-line layout with all text correctly rendered and centered, while Model A breaks “NIGHT MARKET” across two lines. Model A has a polished framed-poster look, but Model B adheres more closely to the prompt and remains crisp and legible. (Second judge pass, order swapped — scores are the average of both: Model B presents the exact three requested lines, cleanly centered on a deep navy, subtly textured background. Model A renders the wording legibly but breaks the | Anthropic: Claude Fable 5.1: Both render all three lines correctly spelled an…
Halimede Palette Alley
A vibrant flat-vector illustration of a narrow city alley after light rain, featuring stacked fruit crates, a noodle-cart awning, overhead wires, a sleeping striped cat on a milk crate, and three pedestrians with umbrellas, composed in clean geometric shapes and restricted strictly to the named Halimede palette: ink black, chalk white, persimmon orange, petrol blue, sage green, and muted lilac, with no stray off-palette colors anywhere; crisp poster-like design, eye-level view, 16:9.
OpenAI: GPT-6 Luna: Model A more clearly includes the requested fruit crates, noodle-cart sign and awning, striped cat on a crate, three umbrella-carrying pedestrians, and rain puddles, while staying closer to the restricted palette. Model B has a more atmospheric, spacious composition, but its cart and fruit details are less explicit and its colors vary more beyond the named palette. (Second judge pass, order swapped — scores are the average of both: Model B creates a more atmospheric, polished alley scene, with c | Anthropic: Claude Fable 5.1: Model A delivers a crisp, bold flat-vector post…
Matchup powered by OpenRouter.