Head to head: Bytedance Seedream V4.5 Edit vs OpenAI: GPT-5.4 Image 2
Bytedance Seedream V4.5 Edit vs OpenAI: GPT-5.4 Image 2
By Ryan Merket · Published
This matchup tests more than visual polish: it asks whether each image model can follow exact counts, object relationships, framing, and distinct illustration styles. The results also expose a recurring trade-off between satisfying a scene brief and honoring its requested aspect ratio.
OpenAI: GPT-5.4 Image 2 leads the saved task sample with an aggregate score of **61.5** to Bytedance Seedream V4.5 Edit’s **49.0**. The task tally is four wins for OpenAI, two for Seedream, and two ties; the editorial assessment is a **slight edge** for OpenAI, not a runaway result. OpenAI’s clearest advantage is instruction fidelity on constrained scenes. It gets the exact seven-cup row and nine-bottle layout right, and correctly places the duck on the green cube. Seedream’s extra cups, excess or cropped bottles, and duck-on-sphere placement are consequential misses, not cosmetic differences. OpenAI also wins the restricted-palette campsite, though its square framing falls short of the requested 16:9. Seedream earns its two wins where framing and graphic treatment matter: it delivers the requested wide perspective in the library scene and the flat-vector, rainy-dawn look for the tea tin. OpenAI’s library shelves are more consistently scaled, and its tea-tin image is polished, but both lose ground against Seedream’s closer match to the specified framing or style. The watchmaker and repair-café scenes are ties: each model has strengths, but neither produces an unqualified match to the brief. OpenAI’s repeated square outputs are a real weakness across several tasks, even when the content is strong. **Verdict: OpenAI: GPT-5.4 Image 2 takes the slight edge on this saved task sample**, chiefly through more reliable counting and object relationships; that assessment applies only to this sample.
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 clear, polished mountain campsite in the requested four-color flat-vector style, though it misses the 16:9 aspect ratio. Model A has the requested wide format, but its gradients violate the palette constraint and oversized foreground trees obscure the campsite. (Second judge pass, order swapped — scores are the average of both: Model B presents a clear, balanced mountain campsite with a tent and campfire, and largely uses the requested four-color palette, though its square for | Anthropic: Claude Fable 5.1: Model A keeps the 16:9 ratio but breaks the cor…
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 many extra cups arranged in additional rows, violating the exact-count and single-row requirements. Model B shows exactly seven separate cups in a straight row, though its square format and the large empty area to the right miss aspects of the requested composition. (Second judge pass, order swapped — scores are the average of both: Model B shows exactly seven clearly separate, identical cups in one straight row on white marble, though its square framing misses the requested 16:9 for | Anthropic: Claude Fable 5.1: Model A shows ten cups arranged in a T-shape (s…
Watchmaker with precision grip
A documentary-style studio photograph of a middle-aged watchmaker seated at a narrow bench, clearly visible from head to waist, carefully holding a tiny brass gear between the thumb and index finger of the left hand while the right hand steadies a loupe near the eye, all ten fingers naturally formed and anatomically correct, realistic forearms and shoulders, neutral posture, a folded microfiber cloth, open parts tray, and compact consumer desk lamp on the bench, crisp side lighting, shallow depth of field, highly detailed skin and fabric textures, 16:9
OpenAI: GPT-6 Luna: Model B better matches the requested head-to-waist documentary view and clearly shows the watchmaker using a loupe while holding a tiny gear, with the bench items and lamp in view. Model A has excellent texture and detail, but its very tight crop omits the full head-to-waist composition and makes the gear and workspace less clear. (Second judge pass, order swapped — scores are the average of both: Model B more fully matches the requested documentary workshop scene, with the watchmaker framed fro | Anthropic: Claude Fable 5.1: Model A follows the brief closely: 16:9 studio…
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 correctly places the yellow duck on the green cube, with the red sphere to its right and the blue cylinder behind both. Model A has the right object colors but places the duck on the red sphere and has a visually confusing, oversized arrangement. (Second judge pass, order swapped — scores are the average of both: Model B correctly places the yellow duck on the green cube, with the red sphere to its right and the blue cylinder behind both. Model A has appealing studio lighting, but puts t | Anthropic: Claude Fable 5.1: Model A places the yellow duck on the red spher…
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 better matches the requested 16:9 framing and warm overhead lighting while establishing a clear centered vanishing point. Model B has especially regular, consistent shelf and book scaling, but its square format misses the explicit aspect ratio. (Second judge pass, order swapped — scores are the average of both: Model B presents a clear, balanced one-point perspective with books and shelves scaling consistently toward the vanishing point, though it misses the requested 16:9 format. Model | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 frame with a…
Pop-up repair café scene
A coherent multi-subject documentary photojournalism scene inside a neighborhood pop-up repair café where a bicycle courier in a cobalt rain jacket hands a scratched wireless speaker to an elderly technician, a teenager tests a refurbished blender at the next table, and a barista in an apron sorts labeled cables into clear bins nearby, folding tables covered with everyday consumer gadgets, handwritten price tags, extension cords, tool rolls, and mugs of tea, balanced triangular composition with all subjects interacting naturally in one space, warm window light mixed with practical bulbs, lively but believable detail, 16:9
OpenAI: GPT-6 Luna: Model B more clearly brings together the specified handoff, blender test, and cable sorting in a balanced, lively repair-café scene, with especially strong details like labeled bins and handwritten tags. Model A is more polished and wide-format, but its subjects and activity feel more spread out, and the courier’s jacket is less clearly cobalt. (Second judge pass, order swapped — scores are the average of both: Model B includes the requested interactions and an abundance of repair-café details, | Anthropic: Claude Fable 5.1: Model B delivers the core narrative more faithfu…
Nine sunscreen bottles
An overhead commercial product photo of exactly 9 distinct sunscreen bottles arranged on pale wet sand after a beach cleanup, every bottle fully visible, separated, and individually countable, with mixed shapes and cap colors but consistent upscale packaging design, tiny shells and a drift line adding context without obscuring any item, bright noon sunlight, sharp shadows, meticulous spacing, high clarity, 16:9
OpenAI: GPT-6 Luna: Model B shows exactly nine distinct, fully visible bottles in a clean overhead layout, with varied packaging, wet sand, and beach debris. Model A has several bottles cropped by the frame and appears to show more than nine, despite its attractive sunlight and product styling. (Second judge pass, order swapped — scores are the average of both: Model B shows exactly nine distinct, fully visible bottles in a clear overhead arrangement, with pale sand, shells, a drift line, and sharp sunlight; its ma | Anthropic: Claude Fable 5.1: Model A contains 10 bottles rather than exactly…
Aurora tea tin illustration
A clean flat vector illustration of a dented mint-green loose-leaf tea tin labeled "Polar Mornings No. 7" sitting on a rainy apartment windowsill beside a tiny cracked ceramic cup, with a wistful late-dawn mood, simplified geometric shapes, subtle paper-grain texture, restrained Nordic palette of sage, apricot, charcoal, and foggy blue, crisp negative space, soft rim light from the window, editorial poster composition, 16:9
OpenAI: GPT-6 Luna: Model A better matches the requested wide, graphic editorial illustration, with the dented mint tin, rainy dawn window, cracked cup, and legible “Polar Mornings No. 7” label. Model B is more polished and balanced, but its near-photographic treatment and square framing depart more from the flat-vector, 16:9 prompt. (Second judge pass, order swapped — scores are the average of both: Model A better captures the requested flat, geometric editorial style, rainy dawn palette, and visibly dented tea ti | Anthropic: Claude Fable 5.1: Model A delivers a true flat vector illustratio…
Matchup powered by OpenRouter.