Head to head: Seedream 5.5 Edit vs Wan
Seedream 5.5 Edit vs Wan
By Ryan Merket · Published
Seedream 5.5 Edit and Wan face off across eight varied image tasks, from strict palette and composition constraints to photorealistic scenes and hands. The comparison turns on different strengths in prompt fidelity, style, and atmosphere.
Seedream 5.5 Edit scored **62.5** and Wan **61.8**, but the saved result is a tie: each took three task wins, with two ties. The editorial strength is **Too close to call (not statistical)**; the small score gap does not change the task-level verdict. Seedream’s clearest showings were the ukiyo-e rain-barrel scene and the golden-hour allotment procession. It captured the requested rainy-night mood with exactly three barrels, and made the six-neighbor sunflower procession read clearly. It also edged Wan on attribute binding, with more natural studio realism and the requested object placement, despite a cube that looked shinier than specified. Wan was stronger on the palette-locked conservatory poster, the bracelet-tying hands, and the freshly watered potting bench. Its flatter graphic treatment better respected the poster’s palette, its hands more clearly conveyed the requested action, and its wet wood, damp terracotta, and leaf droplets sold the just-watered scene. The models split the restricted-palette and perspective tasks: Wan was flatter and more uniform in the former, while Seedream felt more photographic in the latter, but neither separated decisively. This assessment applies to the saved task sample, not to image-model performance in general. **Final call: tie—too close to call, with no overall winner.**
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 requested photorealistic studio scene, with all four objects in the correct positions and colors; Model B also follows the arrangement but looks somewhat more rendered and less natural. Model A has especially strong material detail and balanced lighting, though its cube is a little shinier than the requested matte finish. (Second judge pass, order swapped — scores are the average of both: Model B and Model A both place the duck atop the green cube and keep | Anthropic: Claude Fable 5.1: Both bind colors correctly and place the cube le…
Palette-locked conservatory still life
A meticulous still life inside the tiny rooftop greenhouse of the fictional Saint Ivo Seed Library: five heirloom gourds, a brass mister, a stack of terracotta seed envelopes, and a trailing string-of-pearls plant arranged on a slate shelf, rendered only in the named Hilda’s Garden palette — sage green, dusty rose, marigold yellow, cream, and charcoal — with absolutely no off-palette colors anywhere in the image; flat graphic poster styling with crisp edges, balanced asymmetrical composition, soft overcast daylight, 16:9.
OpenAI: GPT-6 Luna: Model B better follows the palette-locked flat poster style and shows five gourds, while Model A is more polished and atmospheric but leans painterly and appears to include six gourds. Model A’s still life is especially appealing, though Model B adheres more closely to the requested graphic treatment. (Second judge pass, order swapped — scores are the average of both: Model B better honors the strict named palette and flat, crisp poster styling while including the requested still-life elements. | Anthropic: Claude Fable 5.1: Model A delivers a richer, convincingly arranged…
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 A offers a more immersive, balanced campsite scene with polished visual hierarchy, though its subtle shading and color variations stray somewhat from the strict four-color, no-gradient constraint. Model B is more uniformly flat and graphic, but its simplified, object-heavy arrangement feels less cohesive and evocative. (Second judge pass, order swapped — scores are the average of both: Model B follows the flat-vector style and four-color constraint more closely while clearly depicting a mo | Anthropic: Claude Fable 5.1: Model A delivers a cohesive, well-balanced moun…
Ukiyo-e rain barrel gardener
A faithful ukiyo-e woodblock print of a night gardener in indigo work clothes trimming shiso beside three cedar rain barrels behind the fictional Kenzan Tea House, with curved rooflines, patterned waves in puddles, stylized maple leaves blowing across the scene, visible woodblock texture, simplified planes, bold contour lines, and the elegant compositional rhythm characteristic of Edo-period prints; cool evening light after rain, wide view, 16:9.
OpenAI: GPT-6 Luna: Model A captures the cool post-rain night, exactly three cedar barrels, and the requested gardener, shiso, maple leaves, and patterned puddles in a richly textured ukiyo-e-inspired scene. Model B has flatter, more graphic woodblock styling and legible tea-house signage, but its brighter daylight mood and four barrels depart from the prompt. (Second judge pass, order swapped — scores are the average of both: Model A better captures the requested night-after-rain atmosphere, with three cedar barre | Anthropic: Claude Fable 5.1: Model A delivers the night setting, cool post-r…
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 A more clearly depicts a person tying the bracelet onto a wrist, with a natural candid setting and plausible hand positions. Model B is cleanly composed and detailed, but the hands appear to be inspecting or holding an already braided bracelet rather than tying it. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows both hands working the loose threads, with a plausible grip and well-defined, natural-looking fingers. Model A has a warmer candid fe | Anthropic: Claude Fable 5.1: Model A delivers a convincing candid two-person…
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 feels more photorealistic, with convincing variation and perspective scaling in the books while maintaining a clear single vanishing point. Model B has especially strong symmetry and warm lighting, but its repeated, overly uniform details make the scene feel more artificial. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the requested 16:9 framing, warm overhead lighting, and crisp, centered one-point perspective with consistent scale rec | Anthropic: Claude Fable 5.1: Model A delivers a convincingly photorealistic…
Golden-hour allotment procession
A whimsical claymation-style scene of six neighbors carrying absurdly tall sunflower seedlings through Plot 47 of the Bramble Fen allotments, each plant tied with striped twine and wobbling in the breeze, wheelbarrows half-buried in nasturtiums, a crooked shed painted with hand-lettered signs in the background, all bathed in rich golden-hour light with long amber shadows, glowing rim light, and a warm late-summer mood, cinematic wide composition, 16:9.
OpenAI: GPT-6 Luna: Model A more clearly delivers the six-neighbor procession with towering, twine-tied sunflowers, nasturtium-filled wheelbarrows, and a Plot 47 shed in rich golden-hour light. Model B is charming and detailed, but its procession is less clear and the busy signage and staging weaken the cinematic composition. (Second judge pass, order swapped — scores are the average of both: Model A clearly presents six neighbors carrying tall, twine-tied sunflowers in a cinematic golden-hour procession, with nast | Anthropic: Claude Fable 5.1: Model A delivers all six neighbors hoisting abs…
Photoreal potting bench after watering
A photorealistic close-to-mid shot of a cedar potting bench in a real backyard nursery moments after watering: damp terracotta pots, a galvanized steel trowel, folded indigo canvas gloves, a half-zipped bag of bark mix, droplets clinging to basil leaves, and the gardener’s forearm resting on the bench with believable skin texture; diffuse morning light from the left creates physically accurate reflections, soft contact shadows, darkened wet wood grain, matte soil, translucent water beads, and convincing material separation throughout, 16:9.
OpenAI: GPT-6 Luna: Model B presents the requested freshly watered setup with especially convincing wet wood, droplets, and a naturally resting forearm, while its tighter framing gives the objects clear presence. Model A includes the same key elements, but the curled hand and busier arrangement feel slightly less natural. (Second judge pass, order swapped — scores are the average of both: Model B more directly captures the freshly watered scene, with conspicuously wet wood, damp terracotta, leaf droplets, and the g | Anthropic: Claude Fable 5.1: Model B more convincingly sells the 'moments af…
Matchup powered by OpenRouter.