Head to head: Bytedance Seedream V4.5 Edit vs Krea 3 Edit
Bytedance Seedream V4.5 Edit vs Krea 3 Edit
By Ryan Merket · Published
A head-to-head across image-editing prompts that test precision, composition, text rendering, and style fidelity. The results show how each model handles very different kinds of constraints.
Krea 3 Edit takes the saved verdict, scoring **64.6** to Bytedance Seedream V4.5 Edit’s **53.6**. It won five tasks to Seedream’s two, with one tie: a clear result on this task sample, not a statistical claim. Krea’s edge came on prompts where exact execution matters. It showed seven countable cups where Seedream showed six and cropped the row; it rendered all three requested text lines legibly where Seedream’s oversized lettering was cut off; and it more clearly handled the specified object relationships and isometric room layout. Its ukiyo-e wave also made the named style more convincing. Seedream was credited with wins on reflections and glass, and perspective and scale. The restricted-palette campsite was a tie, so Krea’s advantage was broad rather than a sweep. Still, those two Seedream wins were not enough to offset Krea’s stronger showing on counting, typography, layout, and style adherence. This assessment applies to the saved task sample only; it should not be read as a general claim about either model. **Verdict: Krea 3 Edit wins this head-to-head on more consistent execution of the brief.**
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 has a polished overhead look, but shows only six cups and crops the row at the right edge. Model B clearly presents exactly seven individually countable cups in a straight row, though they are small within the frame. (Second judge pass, order swapped — scores are the average of both: Model B shows exactly seven separate, identical espresso cups in a straight overhead row on white marble. Model A has fewer than seven visible cups and crops the row at the edges, despite its polished appear | Anthropic: Claude Fable 5.1: Model A shows only six cups, failing the core '…
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 A uses the requested wording and navy background, but the oversized text overlaps and is severely cropped, making the lines illegible. Model B presents all three lines crisply, correctly spelled, centered, and legible in a clean minimalist layout. (Second judge pass, order swapped — scores are the average of both: Model B presents all three lines of correctly spelled text clearly, centered on a deep navy background in a restrained sans-serif style. Model A uses oversized, heavily cropped t | Anthropic: Claude Fable 5.1: Model A renders all text correctly and crisply…
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 has a more coherent, photorealistic arrangement, with the chrome teapot, water glass, marble, window, and red apple all clearly present; reflections are visible, though the apple and window are not unmistakably reflected on the water’s surface. Model A delivers more dramatic lighting and a stronger close-up, but its oversized, cropped objects and less natural-looking details weaken the overall result. (Second judge pass, order swapped — scores are the average of both: Model A more clearl | Anthropic: Claude Fable 5.1: Model A clearly shows the window and red apple…
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 B presents the requested object relationships clearly, with the duck sitting on the cube and the cylinder behind the foreground objects. Model A is more polished and photorealistic, but its oversized duck and tight framing make the arrangement less balanced. (Second judge pass, order swapped — scores are the average of both: Model B clearly shows all four objects with the duck on top of the green cube, the red sphere to its right, and the blue cylinder behind them. Model A is polished, but | Anthropic: Claude Fable 5.1: Both place a green cube left of a red sphere wi…
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 clear, balanced flat-vector campsite scene with mountains, dusk details, and a cohesive restricted palette. Model A includes the required elements and palette, but the oversized, severely cropped tent and textured, less-flat rendering weaken the composition and style match. (Second judge pass, order swapped — scores are the average of both: Model B clearly presents a balanced mountain campsite at dusk in a clean flat-vector style, with the tent, fire, and layered mountains all | Anthropic: Claude Fable 5.1: Model A delivers a cleaner retro flat-vector mo…
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 presents a clearer, more geometrically consistent one-point-perspective aisle, with books and shelves receding naturally toward a centered vanishing point. Model A has appealing warm light and strong depth, but its pronounced blur and less regular shelf geometry weaken the requested sense of realistic, undistorted scale. (Second judge pass, order swapped — scores are the average of both: Model B delivers a crisp, balanced one-point-perspective aisle with consistently receding shelves and | Anthropic: Claude Fable 5.1: Model A delivers a convincingly photorealistic…
Spatial layout
A clean isometric illustration of a bedroom: a bed against the LEFT wall, a round rug centered on the floor, a desk under the WINDOW on the back wall, and a floor lamp in the FRONT-RIGHT corner. Flat-vector style, consistent perspective.
OpenAI: GPT-6 Luna: Model B clearly uses a consistent isometric perspective and places the bed, centered round rug, desk beneath the back-wall window, and floor lamp in the requested locations. Model A includes the same furnishings but is presented nearly straight-on rather than isometrically, and its oversized, cropped framing weakens the composition. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the requested clean isometric flat-vector room and places all four furnis | Anthropic: Claude Fable 5.1: Model A places all four items correctly (bed le…
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 convincingly captures ukiyo-e through its flat indigo-and-cream planes, bold contours, and stylized foam, while showing a fishing boat amid a dramatic cresting wave. Model A has an appealing dawn scene and forceful surf, but its sailing vessel and more painterly, softly shaded treatment are less faithful to the prompt. (Second judge pass, order swapped — scores are the average of both: Model B closely captures the ukiyo-e woodblock tradition with a dramatic cresting wave, a small fi | Anthropic: Claude Fable 5.1: Model B delivers a genuinely woodblock-like ima…
Matchup powered by OpenRouter.