Head to head: Wan vs Krea 3 Edit
Wan vs Krea 3 Edit
By Ryan Merket · Published
Wan and Krea 3 Edit are tested across a varied set of image tasks, from precise object placement and exact counts to perspective, reflections, and scene-specific actions. The comparison rewards both visual polish and fidelity to the prompt’s concrete details.
Krea 3 Edit leads the saved results, scoring **63.1** to Wan’s **58.4** and taking six tasks to Wan’s two, with no ties. That makes this a **clear result on this task sample**—not a claim about statistical confidence or performance beyond it. Krea’s edge comes from repeatedly making requested details legible: it better conveyed the anxious curator in a flat-file drawing archive, showed the clockmaker’s caliper meeting a brass baluster, and made the window and apple reflections clearer in the chrome. It also won the saved verdicts for attribute binding, perspective and scale, and the restricted-palette campsite. Wan’s two wins were notably concrete. It delivered exactly eleven clearly separated cobalt vases, where Krea appeared to show twelve, and produced the more unified one-point perspective in the reading room, including the near chair and ladder. Those are meaningful wins, but they did not outweigh Krea’s broader set of task-level advantages. **Verdict: Krea 3 Edit wins this head-to-head on the saved task sample, with a clear result.** Wan has standout precision on counting and one perspective scene; Krea was more consistently faithful to the requested composition and action across the other tasks.
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 more convincingly delivers the photorealistic tabletop look, with natural wood grain and soft studio lighting, while keeping the duck on the green cube and the blue cylinder behind the objects. Model B follows the requested object layout clearly, but looks more like a clean CGI render and is less visually rich. (Second judge pass, order swapped — scores are the average of both: Model B more faithfully matches the requested matte green cube, glossy red sphere, and blue cylinder without ad | Anthropic: Claude Fable 5.1: Model B places the small blue cylinder clearly…
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 A has a clear, bright one-point perspective and warm overhead lighting, while Model B’s stronger symmetry, convincing depth, and rich library detail make the aisle feel more immersive. Both keep the aisle empty and show consistent scale receding toward the vanishing point. (Second judge pass, order swapped — scores are the average of both: Model A has especially clear, balanced one-point perspective, with shelves and books consistently shrinking toward the centered vanishing point. Model B | Anthropic: Claude Fable 5.1: Both deliver a convincing one-point perspective…
Anxious curator in archive hall
A cinematic 16:9 editorial photograph of a young museum curator standing alone in the echoing central aisle of the Vantrel Municipal Drawing Archive, visibly trying to hide sharp anxiety before an inspection: tight jaw, raised inner eyebrows, glassy eyes, one forced half-smile that doesn’t reach the face, shoulders subtly tensed, fingers gripping a rolled floor plan too hard; behind her, towering flat-file cabinets and suspended pendant lamps recede into soft depth, with cool morning skylight mixed with dim amber practicals, realistic skin texture, shallow depth of field, emotionally unmistakable expression.
OpenAI: GPT-6 Luna: Model B more clearly depicts the flat-file archive and makes the curator’s anxiety unmistakable, with the rolled plan and inspection-ready central aisle. Model A is more polished and formally dressed, but its shelving reads more like a library than a drawing archive. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the drawing archive with its towering flat-file cabinets and shows anxiety vividly, with both hands gripping the rolled plan. Model A h | Anthropic: Claude Fable 5.1: Model B nails the Drawing Archive setting with…
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 captures the close-up and shows the window and red apple reflected in the chrome, while Model A has a pleasing, balanced still life and especially clear countertop reflections. Neither makes the apple reflection on the water’s surface clearly legible, though Model B’s glass shows stronger reflective detail overall. (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 sid | Anthropic: Claude Fable 5.1: Model B clearly shows both the window grid and…
Reading room vanishing lines
A minimalist line art interior rendering of the newly finished Kestrin Ferry Reading Room, viewed from one corner so the oak floorboards, ceiling beams, book alcoves, and long communal tables all converge to a single consistent vanishing point; include a near foreground chair, midground ladder, and distant circulation desk with correct relative scale so every object shrinks naturally with distance, no warped geometry, no floating edges, bright noon light through clerestory windows, ultra-clean architectural composition, 16:9.
OpenAI: GPT-6 Luna: Model A establishes a clear, consistent single vanishing point and includes the near chair, ladder, communal table, bookshelves, beams, and distant desk with convincing scale. Model B includes brighter clerestory windows, but its near chair is missing, the ladder is awkwardly rendered, and the perspective feels less unified. (Second judge pass, order swapped — scores are the average of both: Model B better suggests a corner viewpoint and includes bright high windows, but its ladder and perspecti | Anthropic: Claude Fable 5.1: Model A delivers a crisp, bold-lined one-point…
Eleven cobalt vases in atrium
A sunlit contemporary hotel atrium photographed in a crisp design-magazine style, featuring EXACTLY eleven distinct matte cobalt ceramic vases arranged across a pale travertine reception ledge—each vase fully visible, separated enough to count individually, differing slightly in silhouette but clearly the same material and color; no extra vessels anywhere else in the scene, with soft shadows from a skylight, calm neutral interior palette, and a balanced 16:9 composition.
OpenAI: GPT-6 Luna: Model A shows exactly eleven clearly separated cobalt vases on a pale stone ledge in a bright atrium, while Model B appears to show twelve, missing the prompt’s exact-count requirement. Model B offers more varied silhouettes, but Model A’s count accuracy and balanced skylit composition make it the stronger match overall. (Second judge pass, order swapped — scores are the average of both: Model B has appealing cobalt forms but appears to show twelve vases rather than exactly eleven, and its large | Anthropic: Claude Fable 5.1: Model A delivers exactly eleven clearly countab…
Clockmaker measuring stair model
A highly detailed realist studio portrait of an elderly clockmaker-architect in a narrow workshop, shown from knees up beside a handmade walnut staircase maquette, both hands fully visible as he carefully measures a brass baluster with a caliper—clear natural anatomy, five distinct fingers on each hand, believable knuckles, tendons, wrists, and relaxed posture, with accurate body proportions; shelves of gears, drafting templates, and labeled wood samples fill the background, lit by crisp side window light and a small task lamp, composed cleanly in 16:9.
OpenAI: GPT-6 Luna: Model B makes the caliper visibly meet a brass baluster, matching the central action especially clearly, while keeping the elderly craftsperson and staircase maquette prominent. Model A has a richly detailed workshop and an impressive model, but the measuring action is less legible. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows the caliper measuring a brass baluster on the walnut staircase maquette, with the requested window light and task lamp. M | Anthropic: Claude Fable 5.1: Model B better honors the specifics—a walnut ma…
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 presents a cleaner, more balanced campsite scene with a clear dusk sun, mountains, tent, and campfire; Model A includes more camping details but feels cluttered and has an awkward central flag-and-smoke shape. Both largely use the requested four-color palette, though Model B’s simpler, crisp vector treatment adheres more convincingly overall. (Second judge pass, order swapped — scores are the average of both: Model B more clearly conveys dusk with the setting sun and crescent moon, and i | Anthropic: Claude Fable 5.1: Model A sticks very strictly to the four-color…
Matchup powered by OpenRouter.