Head to head: Seedream 5.5 Edit vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
Seedream 5.5 Edit vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
By Ryan Merket · Published
A close matchup across boutique scenes, reflective surfaces, precise layouts, textile detail, typography, and constraint-following. The outputs reward different priorities: editorial polish and atmosphere on one side, exact spatial compliance on the other.
Seedream 5.5 Edit leads the saved task sample with an aggregate score of **62.8 to 60.5**, winning **6 tasks to Nano Banana 2’s 1, with 1 tie**. The edge is built on a broader set of stronger results, not on flawless execution: both models missed parts of the counted boot-display brief, and the reflection task exposed a tradeoff between dramatic lighting and legibility. Seedream’s clearest advantages came where finish and visual detail mattered. It was favored for the macro visor portrait’s intimate framing and glossy textures, the jacket’s crane jacquard and stitch detail, and the minimalist poster’s crisp, correctly spelled text. Its reading nook also better preserved the cozy composition, though greenery outside the window weakened the strict no-plants constraint for both models. Nano Banana 2’s sole task win was the precise accessory layout: it kept the items separated and the receipt visible, while Seedream’s receipt was cropped. The models tied on attribute binding; both got the main object colors and placements right, with different staging drawbacks. These exceptions matter, but they do not outweigh Seedream’s stronger run across the remaining tasks. **Verdict: Seedream 5.5 Edit is the saved winner, and this is a decisive editorial result for this task sample.** That judgment applies only to the saved task sample, not to image editing in general.
Counted boutique display
A hyperreal editorial photo of a fashion boutique window display at blue hour, showing exactly 11 distinct ankle boots on separate clear acrylic pedestals: 3 cherry-red patent leather, 2 matte ivory suede, 4 gunmetal snakeskin, and 2 translucent smoky PVC, every boot fully visible and individually countable with no overlaps or reflections hiding any pairmates; polished concrete floor, faint neon signage glow, crisp studio-quality lighting, symmetrical composition, 16:9.
OpenAI: GPT-6 Luna: Model A has a more polished editorial look, with strong blue-hour atmosphere, crisp materials, and an elegant display; however, it appears to show only 10 boots, including just one smoky PVC boot. Model B also appears to have 10 boots and misses the requested color count, while its central mullion and busier framing make the display less clear. (Second judge pass, order swapped — scores are the average of both: Model A has a more polished, balanced boutique-window composition and more clearly ma | Anthropic: Claude Fable 5.1: Both outputs fall short of the required 11 boot…
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 A has lovely dramatic side light and a clearly visible apple, but the required apple reflection in the water is difficult to discern. Model B makes the window and apple reflections much more legible across the chrome and glass, though its lighting is less dramatic and the apple’s position through the glass is somewhat ambiguous. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows the requested window and red-apple reflections in the chrome and gla | Anthropic: Claude Fable 5.1: Model A delivers a convincingly photorealistic…
Macro visor portrait
A cinematic cyberpunk fashion portrait of a young model wearing a translucent rain hood and mirrored visor, shot as an extreme close-up macro with a 100mm lens look, razor focus on the water droplets clinging to the visor rim and the fine texture of glossy vinyl lipstick, while the ears, hood edge, and neon kanji signs behind fall into creamy shallow bokeh; magenta and cyan rim lighting, off-center composition, moody night atmosphere, 16:9.
OpenAI: GPT-6 Luna: Model A better delivers the extreme macro feel, glossy lipstick detail, water droplets, shallow bokeh, and dramatic cyan-magenta lighting. Model B has the more clearly mirrored visor and recognizable neon signage, but its framing is less intimate and the background is more defined than the prompt suggests. (Second judge pass, order swapped — scores are the average of both: Model A better captures the extreme macro framing, crisp glossy lipstick texture, droplets, shallow bokeh, and magenta-cyan | Anthropic: Claude Fable 5.1: Model A nails the extreme macro feel, glossy vin…
Precise accessory layout
A top-down flat-lay product photograph of a fashion stylist’s workstation on pale gray stone, arranged with exact spatial relationships: a cobalt silk scarf lies in the center; a silver safety pin is on top of the scarf near its upper-right corner; a pair of tortoiseshell sunglasses is directly left of the scarf; a coral lipstick is directly right of the scarf; a black measuring tape is above the scarf; a folded cream receipt is below the scarf; and a single pearl earring is precisely between the sunglasses and the scarf; clean softbox lighting, minimalist editorial styling, every object fully visible and clearly separated, 16:9.
OpenAI: GPT-6 Luna: Model B follows the requested spatial arrangement more precisely, with every accessory clearly separated and the receipt fully visible; Model A has an attractive editorial look but the receipt is cropped and the scarf spreads broadly across the layout. Model A’s patterned scarf is visually appealing, while Model B’s simpler styling better supports the prompt’s exact-layout emphasis. (Second judge pass, order swapped — scores are the average of both: Model B follows the requested spatial layout e | Anthropic: Claude Fable 5.1: Model B nails the exact spatial brief: solid co…
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 cleanly matches the specified object colors and placements, with a convincing matte cube, glossy sphere, and duck on top; its soft studio setup is somewhat undercut by a strong shadow and distracting dark shape in the background. Model B also gets the main arrangement right, but its tabletop setting feels less like a studio scene and its duck is oversized relative to the cube. (Second judge pass, order swapped — scores are the average of both: Model B cleanly shows all requested objects | Anthropic: Claude Fable 5.1: Both bind colors correctly with the cube left, g…
Jacquard precision jacket
An ultra-detailed fashion studio portrait of a cropped tailored jacket draped on a dress form, rendered in contemporary luxury editorial style, featuring intricate midnight-teal jacquard weave with tiny copper-thread cranes, sharply resolved herringbone silk lapels, hand-finished topstitching, brushed brass zipper teeth, and a soft cashmere lining peeking from one cuff; every thread, weave transition, and seam texture must appear crisp and unsmudged under controlled side lighting against a dark neutral backdrop, 16:9.
OpenAI: GPT-6 Luna: Model A more closely captures the requested dark luxury editorial look, with clearly resolved herringbone lapels, copper cranes, visible topstitching, brass zipper, and a cuff lining. Model B has a clean, balanced presentation, but its brighter fabric and less distinct lapel and weave details are less faithful to the prompt. (Second judge pass, order swapped — scores are the average of both: Model A more fully delivers the requested luxury textile detail, with sharply visible crane jacquard, her | Anthropic: Claude Fable 5.1: Model A delivers a macro-level editorial shot w…
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 more directly follows the deep-navy, minimalist poster brief, with all three lines crisp, correctly spelled, and centered. Model B renders the text well too, but the off-white mockup border makes it less faithful to the requested background. (Second judge pass, order swapped — scores are the average of both: Model B renders all three lines correctly and has a restrained poster treatment, but the surrounding mockup and generous empty space make it feel less direct. Model A presents the cr | Anthropic: Claude Fable 5.1: Model A renders all three lines perfectly spell…
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 better captures the cozy, warmly lit reading nook and keeps the chair, books, and steaming tea clearly composed by the window; Model B also includes all requested furnishings but has prominent greenery visible through the window and distracting pseudo-text on the book spines. Both avoid lamps and wall art, though greenery outside the windows weakens the strict no-plants requirement in both images. (Second judge pass, order swapped — scores are the average of both: Model B better matches | Anthropic: Claude Fable 5.1: Both Model A and Model B honor the negations ind…
Matchup powered by OpenRouter.