Head to head: Wan vs Krea 3 Edit
Wan vs Krea 3 Edit
By Ryan Merket · Published
Wan and Krea 3 Edit face off across eight image tasks spanning exact counts, constrained palettes, named styles, and detailed scene construction. The comparison turns on how consistently each model follows specific visual instructions while keeping its images coherent.
Krea 3 Edit takes five task wins to Wan’s three, with no ties, and finishes at 63.4 against Wan’s 59.6. The gap is modest: the saved editorial assessment is a **slight edge**, not a decisive separation. Krea’s strongest results come in scenes where several prompt details need to work together. Its assay bench includes the requested stool, apron, task lamp, and monitor spill; its empty test track makes the calibration gear unmistakable; and it gives the campsite and fishing-boat images cleaner, more coherent compositions. It also wins the attribute-binding task, though the judges noted that both images had imperfections. Wan is better in three telling cases: the Bauhaus classroom feels more distinctly like a demonstration, its overhead image gets the espresso-cup count right while showing coffee in the cups, and its cryolab still life has the more cohesive workstation arrangement. Those wins show why Krea’s lead is narrow rather than a blanket verdict on every kind of prompt. **Verdict: Krea 3 Edit, by a slight edge.** It wins more tasks and scores higher overall, but Wan has clear strengths in exact-count presentation and certain art-directed scenes. This assessment applies to the saved task sample.
Bauhaus isotope poster scene
A faithful Bauhaus-style poster-like illustration of a fictional isotope chemistry classroom demonstration, 16:9: a lecturer in simplified geometric form stands beside a blackboard marked 'Xe-129 Lab', while three students sit at modular desks with beakers and measuring rods; use strict Bauhaus visual language—flat primary colors plus black and cream, bold sans-serif typography, circles/triangles/rectangles, asymmetrical grid composition, minimal shading, strong negative space, and disciplined graphic balance—so the image unmistakably reads as Bauhaus rather than generic modernist design.
OpenAI: GPT-6 Luna: Model A more fully delivers the chemistry classroom scene, with a lecturer, three students, lab equipment, and bold geometric Bauhaus poster styling. Model B has a clean abstract layout and the requested blackboard text, but its thin typography and less legible figures make it feel less distinctly Bauhaus and less like a classroom demonstration. (Second judge pass, order swapped — scores are the average of both: Model A more unmistakably evokes Bauhaus through its bold geometric figures, primary | Anthropic: Claude Fable 5.1: Model A delivers a dense, authentic Bauhaus pos…
Tungsten-lit assay bench realism
A photorealistic close-to-mid shot of a late-night materials-science lab bench illuminated by a single tungsten task lamp and cool spill from a monitor, 16:9: a brushed stainless microscope, a varnished maple stool, a folded navy cotton lab apron, a researcher’s hand with natural skin texture holding ceramic tweezers over a petri dish of iridescent thin-film samples, and a scratched borosilicate beaker half-filled with amber liquid; emphasize physically convincing light falloff, soft and hard shadow interplay, subtle reflections, fingerprints on metal, wood grain, fabric weave, translucent glass thickness, and believable specular highlights without any CGI glossiness.
OpenAI: GPT-6 Luna: Model B includes nearly every requested element, especially the varnished maple stool, navy apron, tungsten task lamp, and cool monitor spill, with convincing bench textures and light falloff. Model A is polished and clearly shows the hand, iridescent samples, and amber beaker, but substitutes a wooden board for the stool and feels more staged. (Second judge pass, order swapped — scores are the average of both: Model B includes more of the requested scene details, especially the varnished maple | Anthropic: Claude Fable 5.1: Model A has a cleaner composition with a clearly…
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: Model A shows exactly seven clearly separated, matching espresso cups in a straight overhead row, with espresso visible and soft lighting on white marble. Model B also has seven cups, but the empty interiors and washed-out lighting make it a less compelling match. (Second judge pass, order swapped — scores are the average of both: Both images show exactly seven clearly separated cups in a straight row on white marble. Model A has a slight edge because the cups contain espresso, while Model B has | Anthropic: Claude Fable 5.1: Both Model A and Model B render exactly seven s…
Sterile test track with no vehicles
A wide-angle view of an outdoor autonomous-laboratory proving ground at dawn, 16:9: pale concrete lanes curve past sensor towers, calibration targets, wind socks, safety bollards, and a glass control kiosk marked 'Sector L-7', with two technicians in silver cleanroom jackets checking clipboards beside orange traffic cones; the scene must deliberately contain NO cars, trucks, buses, motorcycles, bicycles, scooters, carts, forklifts, drones, or any other vehicles whatsoever, leaving the expansive test track completely empty of transport machines; photorealistic style, low sun, long shadows, faint ground mist.
OpenAI: GPT-6 Luna: Model B more clearly captures the curved, empty proving ground and includes prominent calibration targets, sensor towers, a windsock, and the correctly marked kiosk. Model A is also atmospheric and adheres well, but its testing equipment is less explicit and the scene feels less like a specialized proving ground. (Second judge pass, order swapped — scores are the average of both: Model B more closely captures the requested sweeping curved lanes, calibration targets, sensor towers, and empty prov | Anthropic: Claude Fable 5.1: Model B satisfies more of the prompt's specific…
Attribute binding
A photorealistic tabletop scene: a matte green cube to the LEFT of a glossy red sphere, a small blue cylinder BEHIND both, and a single yellow rubber duck sitting ON TOP of the cube. Colors must stay on the correct objects. Soft studio light, 16:9.
OpenAI: GPT-6 Luna: Model A better matches the photorealistic look and clearly places the duck atop the cube with the cylinder behind the foreground objects, though the dark intrusion at the upper left is distracting. Model B has a cleaner layout, but looks more like a vivid render and has a small odd artifact on the sphere. (Second judge pass, order swapped — scores are the average of both: Model B presents all four objects clearly in the requested arrangement with a clean studio composition. Model A also follows | Anthropic: Claude Fable 5.1: Both bind colors correctly and place the duck on…
Palette-locked cryolab still life
A meticulously arranged still life inside a compact cryobiology workstation, shown in a clean three-quarter view, 16:9: three labeled nitrogen dewars, a frost-rimmed specimen rack holding 11 glass ampoules, a hexagonal timer, insulated tongs, and a clipboard with tiny diagrams, rendered as a pastel storybook illustration with crisp edges and soft grain; restrict the entire image ONLY to the named color palette 'Rose Quartz' (#F7CAC9), 'Serenity' (#92A8D1), 'Cream Gold' (#F2E6B3), 'Warm Taupe' (#B89C8A), and 'Charcoal Ink' (#2F3640), with no other hues anywhere in the image, including highlights, shadows, reflections, labels, and background.
OpenAI: GPT-6 Luna: Model A better captures the compact, carefully arranged cryolab still life, with a frosted rack, timer, tongs, clipboard diagrams, and a cohesive pastel illustration; however, it shows only two dewars and has some garbled text. Model B includes three dewars but the labels do not identify nitrogen, the rack lacks a clear frost rim, and the sparse, pale setting feels less like a compact workstation. (Second judge pass, order swapped — scores are the average of both: Model A has the more cohesive p | Anthropic: Claude Fable 5.1: Model A shows only two dewars and has garbled l…
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 delivers a cleaner, more balanced flat-vector campsite scene and uses the four-color palette consistently, though its moonlit sky reads more as night than dusk. Model A suggests dusk more clearly but feels cluttered, with an awkwardly crowded arrangement of campsite objects. (Second judge pass, order swapped — scores are the average of both: Model B more cleanly follows the four-color, flat-vector constraint and presents a clear dusk campsite with a balanced mountain-and-foreground compo | Anthropic: Claude Fable 5.1: Model B holds the four-color palette cleanly wi…
Named art style
A ukiyo-e woodblock print of a fishing boat riding a large cresting wave at dawn, faithful to the flat color planes, bold outlines, and stylized foam of the tradition, muted indigo and cream palette.
OpenAI: GPT-6 Luna: Model B more clearly depicts a single fishing boat riding the wave, with a dawn horizon and restrained indigo-and-cream palette. Model A has especially bold outlines and stylized foam, but its boats feel fragmented and make the scene less coherent. (Second judge pass, order swapped — scores are the average of both: Model B clearly depicts a crewed fishing boat riding the cresting wave and closely follows the muted indigo-and-cream woodblock-print brief. Model A has striking traditional styling, | Anthropic: Claude Fable 5.1: Model A copies Hokusai's layout convincingly but…
Matchup powered by OpenRouter.