Head to head: Bytedance Seedream V4.5 Edit vs Google: Nano Banana Pro (Gemini 3 Pro Image)
Bytedance Seedream V4.5 Edit vs Google: Nano Banana Pro (Gemini 3 Pro Image)
By Ryan Merket · Published
Eight image-editing prompts put Seedream V4.5 Edit and Nano Banana Pro through layout, object placement, style, and explicit-constraint tests. The results turn on how consistently each model handled the full brief, not just whether it produced an attractive image.
Nano Banana Pro takes the saved matchup, winning **7 of 8 tasks** with no ties. Its aggregate score was **64.9**, against **53.4** for Seedream V4.5 Edit—a decisive result on this task sample, not a statistical claim. The difference showed up in both composition and fidelity. Nano Banana better matched the requested isometric flat-vector bedroom, kept the midnight tea tin, cup, and cardamom pods visible in a coherent poster scene, and handled the Aegean still life and ukiyo-e wave with clearer spacing and more faithful styling. In the flood-cleanup scene, it also showed neighbors at work without introducing a vehicle, which Seedream appeared to do. Seedream’s sole win was the negation prompt: it created the warm reading nook while avoiding plants, lamps, and artwork. Nano Banana let garden foliage show through the window, violating the explicit no-plants constraint. That is a meaningful miss, but it does not outweigh its more consistent execution across the other seven prompts. **Verdict: Nano Banana Pro wins this head-to-head, decisively on the saved task sample.** This assessment applies to these eight tasks only; it should not be read as a claim about performance on every image-editing brief.
Spatial layout
A clean isometric illustration of a bedroom: a bed against the LEFT wall, a round rug centered on the floor, a desk under the WINDOW on the back wall, and a floor lamp in the FRONT-RIGHT corner. Flat-vector style, consistent perspective.
OpenAI: GPT-6 Luna: Model B clearly delivers the requested isometric, flat-vector bedroom and places the bed, centered round rug, desk beneath the back-wall window, and floor lamp in the room. Model A includes the key furnishings but uses a mostly straight-on, more realistic view and feels less faithful to the requested style. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the requested isometric flat-vector style and places the bed, centered round rug, desk beneath the | Anthropic: Claude Fable 5.1: Model A places every object correctly (bed left,…
Car-free flood cleanup
Documentary photojournalism scene of a narrow residential lane just after a burst washing-machine hose flooded several ground-floor flats, with neighbors stacking warped cookbooks, a toppled laundry basket, buckets, mops, and a floral armchair set outside on the wet pavement, overcast morning light, puddle reflections, candid human activity, and absolutely no cars, vans, bicycles, scooters, or any other vehicles anywhere in frame, 16:9
OpenAI: GPT-6 Luna: Model B better captures candid neighbor activity in a flooded residential lane and clearly avoids vehicles. Model A includes many requested cleanup objects, but has no people and appears to show a vehicle in the distant lane. (Second judge pass, order swapped — scores are the average of both: Model B captures candid neighbor cleanup in a wet residential lane, with books, buckets, mops, and a floral armchair and no visible vehicles. Model A has convincing flood-cleanup props and strong puddle ref | Anthropic: Claude Fable 5.1: Model A violates the hard constraint with visib…
Midnight tea tin vector
A clean stylized vector illustration of a dented peacock-blue tea tin, a striped porcelain cup, and three cardamom pods on a narrow kitchen windowsill during a sleepless rainy midnight, seen in three-quarter view with crisp geometric shapes, subtle grain-free shading, long amber reflections from a single under-cabinet light, deep indigo exterior beyond the glass, melancholic but orderly mood, poster-like composition, 16:9
OpenAI: GPT-6 Luna: Model B presents the tin, striped cup, and three cardamom pods clearly in a cohesive poster-like vector scene, with rainy midnight blues and warm under-cabinet light. Model A has evocative lighting and atmosphere, but its close crop cuts off the cup and tin and feels less cleanly vector-stylized. (Second judge pass, order swapped — scores are the average of both: Model B presents all requested objects clearly in a cohesive poster-like vector scene, with the dented blue tin, striped cup, three po | Anthropic: Claude Fable 5.1: Model B delivers a genuinely dented peacock-blu…
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 includes the requested nook elements and warm afternoon light while avoiding plants, lamps, and artwork. Model B has a more fully arranged scene, but visible greenery outside the window violates the explicit no-plants requirement. (Second judge pass, order swapped — scores are the average of both: Model B includes visible garden foliage through the window, violating the explicit no-plants constraint. Model A captures the cozy chair, books, tea, window, and warm afternoon light without th | Anthropic: Claude Fable 5.1: Model A satisfies every constraint—armchair, bo…
Workshop textiles macro
An ultra-detailed close-up still life photograph of a household mending basket tipped onto a maple table: herringbone wool scarf, frayed denim cuff, tomato-shaped pincushion, brass thimble with tiny scratches, tangled linen thread, and a partially unspooled bobbin, rendered with tack-sharp fabric weave, nap, stitching, wood pores, and metal wear under cool north-window daylight, shallow but precise depth of field, documentary realism, 16:9
OpenAI: GPT-6 Luna: Model B presents the requested mending materials more clearly, with crisp fabric and wood detail, while Model A has a convincingly tipped basket but leaves several items cropped or soft and leans warmer than the requested cool daylight. Model B’s basket is less clearly tipped, but its overall completeness and sharpness make it the stronger match. (Second judge pass, order swapped — scores are the average of both: Model B renders nearly all the requested mending objects with crisp, tactile detail | Anthropic: Claude Fable 5.1: Model A follows the literal staging better (cle…
Aegean enamel palette
A carefully arranged still life of household enamelware on a pale tabletop — a coffee pot, colander, bread box, measuring spoons, and folded dish towel — lit by soft side daylight in a minimalist studio, composed with generous negative space, restricted strictly to the named Aegean palette: Prussian blue, seafoam green, warm ivory, terracotta, and charcoal, with no other colors visible anywhere in the image, 16:9
OpenAI: GPT-6 Luna: Model B presents all five requested objects clearly in a balanced studio still life, with generous negative space and soft daylight; Model A includes the items and palette but crops several objects tightly and uses much harsher lighting. Model B also keeps the palette more cohesive, though a few neutral metallic details appear in both images. (Second judge pass, order swapped — scores are the average of both: Model B includes the requested household items in a carefully spaced, softly lit still | Anthropic: Claude Fable 5.1: Model B maps all five Aegean palette colors clea…
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 presents all four objects clearly in the requested arrangement, with the duck sitting on the green cube and the blue cylinder visibly behind the other objects. Model A also gets the color and placement bindings mostly right, but its objects are tightly cropped and dominate the frame, making the composition less balanced. (Second judge pass, order swapped — scores are the average of both: Model B presents all four objects clearly in the requested left-to-right and front-to-back arrangemen | Anthropic: Claude Fable 5.1: Both bind every color to the correct object wit…
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 convincingly captures ukiyo-e through its bold contours, flat indigo-and-cream planes, and expressive stylized foam, with a fishing boat and dawn clearly depicted. Model A has an appealing dramatic wave and dawn palette, but its boat feels more modern and the rendering is less faithful to the woodblock tradition. (Second judge pass, order swapped — scores are the average of both: Model B closely matches the requested ukiyo-e look, with bold outlines, stylized foam, a muted indigo-an | Anthropic: Claude Fable 5.1: Model B nails the requested ukiyo-e conventions…
Matchup powered by OpenRouter.