Head to head: Seedream 5.5 Layerize vs Recraft
Seedream 5.5 Layerize vs Recraft
By Ryan Merket · Published
Eight image prompts put these models through detailed scene-building, strict constraints, exact text, named art styles, and counting. The comparison rewards not just attractive images, but how closely each result follows the brief.
Seedream 5.5 Layerize leads the saved score, **65.4 to Recraft’s 45.0**, with task verdicts of **7 wins to 1** and no ties. The result is decisive as an editorial assessment of this task sample—not a statistical claim. Seedream’s edge is prompt fidelity across demanding briefs. It won on the rooftop sax scene, the six-object backstage still life, the plant-free reading nook, the canal ensemble, the Art Deco poster with exact wording, and the ukiyo-e style test. In each, Recraft’s image could be appealing, but missed a defining instruction: omitting objects or performers, adding a prohibited plant, breaking the poster title, or drifting toward photorealism. The reflections-and-glass result also went to Seedream, though it was less clean-cut: its apple sits on the counter as requested, while Recraft makes some reflections more legible. Recraft’s clearest win was exact counting, with seven cups in a straight, easily counted row; Seedream also showed seven, but arranged them diagonally and added distracting props. **Verdict: Seedream 5.5 Layerize wins this matchup on more consistent execution of detailed instructions; Recraft’s standout is clean, precise counting. This assessment applies only to the saved task sample.**
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 emphasizes the window reflected in the chrome and the layered reflections around the water glass, though the apple sits on the raised sill rather than plainly on the counter and neither image makes both requested reflections on the water surface fully convincing. Model A has a strong close-up and places the apple on the counter, but its reflections are less legible overall. (Second judge pass, order swapped — scores are the average of both: Model A more clearly shows the red | Anthropic: Claude Fable 5.1: Model A places the apple on the counter as spec…
Golden hour rooftop sax solo
A cinematic rooftop performance scene at golden hour: a lone baritone saxophonist in a weathered copper suit stands on the edge of a low apartment roof, long amber sunlight raking across the instrument and casting stretched shadows over potted fig trees, antenna wires, and a faded mural that reads only as texture; the mood is wistful and triumphant, dust glowing in the air, shot in a realistic contemporary film still style with warm lens flare and strong directional sunset light, 16:9.
OpenAI: GPT-6 Luna: Model A more fully captures the prompt with a clearly prominent large saxophone, copper suit, potted fig trees, rooftop antennas, textured mural, and rich amber light with glowing dust. Model B has an appealing silhouette and rooftop setting, but its cooler, brighter light and less distinctive instrument make it feel less aligned with the requested cinematic mood. (Second judge pass, order swapped — scores are the average of both: Model A more clearly conveys the baritone sax, copper suit, rooft | Anthropic: Claude Fable 5.1: Model A delivers a far more cinematic golden-ho…
Backstage binding still life
A hyper-detailed backstage still life on a scuffed rehearsal table, photographed in a clean editorial product style: a matte teal trumpet mute, a glossy crimson metronome, a brushed brass capo, a frosted lavender glass slide, a charcoal-black velvet violin case latch, and a pale mint ceramic espresso cup, each object clearly separated and unmistakably keeping its own color and material with no color swapping or bleeding; shallow depth of field, crisp studio realism, soft neutral light, 16:9.
OpenAI: GPT-6 Luna: Model A clearly presents all six requested objects with distinct colors and materials on a scuffed rehearsal table in a polished editorial still-life style. Model B instead centers on a full teal trumpet and omits several specified objects, despite having a believable backstage setting. (Second judge pass, order swapped — scores are the average of both: Model A clearly presents the requested mute, metronome, capo, lavender slide, velvet case latch, and mint cup with distinct materials and colors | Anthropic: Claude Fable 5.1: Model A renders all six objects with the correc…
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 full cozy reading nook with warm afternoon light and avoids all prohibited items. Model B is visually appealing but prominently includes a plant, violating an explicit constraint. (Second judge pass, order swapped — scores are the average of both: Model B has a prominent potted plant, directly violating the explicit constraint, despite its cozy close-up composition. Model A includes the requested chair, books, tea, window, and warm afternoon light while avoiding plants, lamp | Anthropic: Claude Fable 5.1: Model A correctly includes the armchair, book s…
Canal ensemble tableau
A richly staged multi-subject scene in a stylized realist painting: on a narrow floating stage in a city canal, seven performers interact in one coherent composition — a drummer tuning a compact kit, a cellist seated at stage left, two twin singers sharing a single vintage microphone at center, a bassist leaning toward them, a child conductor on an overturned crate, and a masked stagehand passing sheet music from the rear; paper lanterns reflected in the water, audience clustered on stone steps, balanced triangular grouping, clear sightlines between all figures, lively storytelling, twilight blue ambiance, 16:9.
OpenAI: GPT-6 Luna: Model A closely matches the requested painted canal tableau, including the full ensemble, lantern reflections, audience, and masked sheet-music stagehand; its balanced staging and twilight color make the scene especially compelling. Model B has convincing photographic detail and a canal performance, but omits most of the specified performers and interactions and misses the stylized realist painting treatment. (Second judge pass, order swapped — scores are the average of both: Model A captures th | Anthropic: Claude Fable 5.1: Model A delivers a stylized realist painting wi…
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 B shows exactly seven clearly countable cups in a clean, straight row on white marble, matching the requested flat-lay composition more closely. Model A also has seven cups and is visually appealing, but the row is diagonal and extra props make the scene less clean. (Second judge pass, order swapped — scores are the average of both: Model B shows exactly seven clearly separated, nearly identical cups in a straight horizontal row on white marble with clean overhead styling. Model A also has | Anthropic: Claude Fable 5.1: Model A shows seven cups but in a diagonal row…
Velvet Note poster test
A close, front-facing view of an art-deco venue poster mounted in a brass frame in a dim lobby, designed in elegant high-contrast typography and fully legible: the poster must read exactly 'VELVET NOTE' on the first line, 'ONE NIGHT ONLY' on the second line, and 'MARA QUILL' on the third line; rich cream paper, deep indigo ink, sharp readable letters, tasteful ornamental borders, realistic print texture, photographed like a premium editorial interior detail with soft spotlighting.
OpenAI: GPT-6 Luna: Model A closely matches the requested close, front-facing editorial detail, with all three lines rendered exactly and legibly in an elegant Art Deco design. Model B has a convincing lobby setting, but the title is broken across two lines and the poster’s layout and ornamentation are less faithful to the prompt. (Second judge pass, order swapped — scores are the average of both: Model A renders all three lines exactly and clearly, with a strong art-deco design, brass frame, and atmospheric close- | Anthropic: Claude Fable 5.1: Model A delivers a close front-facing brass-fra…
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 A closely captures ukiyo-e’s flat indigo-and-cream planes, bold contours, stylized foam, and a traditional fishing boat beneath a dawn sun. Model B has a dramatic composition, but its photorealistic scenery and modern-looking boat depart from the requested woodblock-print tradition. (Second judge pass, order swapped — scores are the average of both: Model B includes the cresting wave and boat but leans toward a photorealistic scene rather than the requested flat, traditional woodblock look | Anthropic: Claude Fable 5.1: Model A delivers a coherent ukiyo-e woodblock a…
Matchup powered by OpenRouter.