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
We compare the models on eight image-editing and generation briefs that test constraint-following, composition, style, and anatomy. The results turn on which model handles the full scene—not just its most polished details—more reliably.
OpenAI: GPT-5.4 Image 2 takes the matchup, 64.3 to 58.5, winning five tasks to Seedream’s one, with two ties. The limited confidence verdict reflects a clear aggregate edge, not a sweep: Seedream still has a specific strength, and the closest tasks stayed close. The sharpest difference is constraint-following. In the Ljubljana tram-stop brief, GPT-5.4 avoided both umbrellas and the covered shelter that violated the explicit ban, while still delivering the trio, rainy neon setting, and K-17 sign. It also won on the Memphis boutique, one-point perspective, ukiyo-e style, and the fitting-studio layout. Seedream’s sole task win was hands and anatomy. The models tied on the crystal arcade vitrines and the plant-free reading nook, where different strengths in framing and atmosphere kept either result from pulling ahead. Those ties matter: GPT-5.4 was more consistent overall, but not uniformly better at every brief. **Final call: OpenAI: GPT-5.4 Image 2 wins on reliable scene and constraint execution; Seedream V4.5 Edit has a narrow, real advantage in the hands-and-anatomy test.**
No umbrellas at the tram stop
A rainy midnight fashion editorial on a narrow cobalt-lit tram platform in Ljubljana, with three avant-garde models in glossy plum trench coats and silver ankle boots waiting beneath a flickering route sign that reads K-17; wet pavement, drifting mist, and neon reflections everywhere, but the scene must contain absolutely no umbrellas, no parasols, and no covered awnings of any kind; cinematic cyberpunk neon look, low camera angle, crisp rim lighting, shallow depth of field, 16:9.
OpenAI: GPT-6 Luna: Model A delivers striking neon fashion imagery, but the prominent covered tram shelter directly violates the explicit ban on awnings. Model B avoids umbrellas and awnings while matching the trio, outfits, rainy Ljubljana setting, and clearly rendered K-17 sign. (Second judge pass, order swapped — scores are the average of both: Model B closely follows the scene, with three plum-coated models, silver boots, wet neon-lit pavement, and a clearly rendered K-17 sign, while avoiding umbrellas and awni | Anthropic: Claude Fable 5.1: Model A delivers a polished cinematic cyberpunk…
Hands & anatomy
A natural candid photo of a person tying a friendship bracelet, both hands clearly visible with anatomically correct fingers and a plausible grip on the threads, soft daylight, shallow depth of field, 16:9.
OpenAI: GPT-6 Luna: Model B more clearly shows a person actively tying a friendship bracelet with both hands and a plausible grip. Model A has lovely soft daylight and shallow depth of field, but its very tight crop and awkwardly crowded hand arrangement make the action and anatomy less clear. (Second judge pass, order swapped — scores are the average of both: Model B clearly shows both hands using a believable grip to work the bracelet, with natural anatomy and a candid, softly lit setting. Model A has a striking | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 frame with b…
Memphis knitwear boutique
An interior fashion campaign set inside a tiny experimental knitwear boutique designed in unmistakable Memphis style: squiggle patterns, playful geometric laminates, pastel-turquoise and coral color blocking, terrazzo counters, asymmetric shelving, and bold black graphic accents; a model with a sculptural lemon-yellow sweater dress and mismatched earrings poses beside a zigzag mirror and a cobalt totem lamp; bright editorial studio lighting with soft shadows, straight-on composition, high detail, 16:9.
OpenAI: GPT-6 Luna: Model B brings more of the requested boutique into view, including the cobalt totem lamp, zigzag mirror, terrazzo counter, and playful Memphis graphics. Model A has the requested wide format and a polished editorial look, but its tighter crop and turquoise lamp leave fewer of the key details visible. (Second judge pass, order swapped — scores are the average of both: Model B includes nearly all the requested Memphis boutique details, with a full-length pose and a clear cobalt totem lamp, while M | Anthropic: Claude Fable 5.1: Model B delivers a more authentic Memphis bouti…
Crystal arcade vitrines
A luxury streetwear window display inside a glass-roofed arcade at blue hour, featuring a mannequin in an iridescent smoke-gray bomber jacket and pleated vinyl trousers standing behind a curved crystal-clear vitrine; the scene must show accurate reflections in the shop glass, subtle double reflections from layered panes, believable refraction through a suspended glass orb, chrome trim mirroring nearby magenta signage, and rain-slick stone flooring reflecting the whole composition; polished contemporary fashion photography style, precise optics, balanced symmetrical framing, 16:9.
OpenAI: GPT-6 Luna: Model A more fully captures the requested wide, blue-hour arcade scene, with a striking curved crystal vitrine, suspended orb, layered reflections, and rain-slick flooring. Model B has strong symmetry and shows the full outfit, but its square framing and less clearly curved display fit the prompt less closely. (Second judge pass, order swapped — scores are the average of both: Model B presents the complete mannequin and vitrine in a clear, balanced display, with strong arcade reflections, a refr | Anthropic: Claude Fable 5.1: Model A delivers the requested 16:9 frame, a sw…
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 cleaner, more convincing one-point perspective, with consistently shrinking books and a clear central vanishing point; its main miss is the square format instead of 16:9. Model A has attractive warm lighting and a wide frame, but its heavy haze and softer, less precise shelf geometry weaken the perspective-and-scale brief. (Second judge pass, order swapped — scores are the average of both: Model B delivers a clear, balanced one-point-perspective aisle with convincing scale red | Anthropic: Claude Fable 5.1: Model A delivers the 16:9 ratio and warm light…
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 captures the ukiyo-e tradition, with a traditional fishing boat riding the crest, stylized foam, and a restrained indigo-and-cream palette at dawn. Model A has a striking wave and pleasing print texture, but its boat is partly cropped and feels less central to the scene. (Second judge pass, order swapped — scores are the average of both: Model B more fully captures the prompt, with a clear fishing boat riding the crest, stylized foam, and a balanced dawn scene in muted indig | Anthropic: Claude Fable 5.1: Model B delivers a convincing ukiyo-e print wit…
Exact layout in the fitting studio
In a minimalist fashion fitting studio, arrange the scene exactly as follows: a tailor’s mannequin wearing a saffron blazer stands in the center; a low glass table is directly in front of the mannequin; a folded indigo coat lies on top of the table; a pair of white loafers is on the floor under the table; a brass garment rack is to the left of the mannequin with three empty wooden hangers; a full-length mirror is to the right of the mannequin angled slightly inward; a seamstress in a charcoal jumpsuit stands behind the mannequin; and a small scarlet pin cushion sits between the mannequin and the mirror; soft north-window daylight, clean editorial realism, eye-level composition, 16:9.
OpenAI: GPT-6 Luna: Model A follows the requested wide layout closely, including the three hangers, inward-angled mirror, glass table, folded indigo coat, and shoes beneath the table. Model B is neatly arranged and polished, but the shoes sit in front of the table and the pin cushion is elevated on a pedestal rather than placed between the mannequin and mirror at floor level. (Second judge pass, order swapped — scores are the average of both: Model B includes nearly every requested item in the specified arrangement | Anthropic: Claude Fable 5.1: Model B nails nearly every spatial instruction—…
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 captures the cozy chair, stacked books, tea, and warm window light, with no indoor plants, lamps, or artwork; Model B has a clearer full-nook composition but shows more distinct foliage outside the window, conflicting with the strict no-plants instruction. Model A is also especially atmospheric, though its tighter crop makes the setting less fully visible. (Second judge pass, order swapped — scores are the average of both: Model B presents the complete, inviting reading nook with the cha | Anthropic: Claude Fable 5.1: Model A satisfies every constraint: armchair, b…
Matchup powered by OpenRouter.