Head to head: Bytedance Seedream V4.5 Edit vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
Bytedance Seedream V4.5 Edit vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
By Ryan Merket · Published
Eight image-editing and generation tasks put layout accuracy, constraint-following, readable text and style fidelity under the microscope. The matchup tests whether polished-looking images also get the brief’s specifics right.
Nano Banana 2 swept all eight tasks, with a 66.5 aggregate score against Seedream V4.5 Edit’s 51.0. The result was judged decisive at limited confidence, and the spread reflects a repeated advantage in making requested details clear—not just producing attractive images. That showed up in concrete ways: the loading-dock box sat on the blue pallet, the hard hat was easy to locate, and the harbor scene made the five-person team’s roles and equipment more legible. In the workshop still life, all five objects read clearly as a balanced arrangement; in the union-hall scene, the requested poster wording was centered and readable. Seedream often had appealing lighting and detail, but cropping and placement obscured parts of the brief. Nano Banana 2 also handled style and visual constraints more reliably across the set. Its campsite used a cohesive four-color palette, while its fishing scene made the ukiyo-e influence unmistakable through flat planes, bold outlines and stylized foam. The negation task was the least straightforward comparison, but the overall task verdict still went to Google. **Verdict: Nano Banana 2 wins decisively, 8–0. Seedream V4.5 Edit can deliver polished imagery, but here it too often let composition or stylistic choices get in the way of exact execution.**
Exact loading dock layout
A realistic editorial illustration of an early-morning bakery loading dock, viewed from a slightly elevated wide angle in cool dawn light, 16:9: a red hand truck stands to the left of a stack of flour sacks, a silver milk can is behind the flour sacks, a taped cardboard box sits on top of a blue pallet, and a yellow hard hat lies on the ground exactly between the hand truck and the pallet; the arrangement must be precise, clean, and unambiguous, with workers’ tools and painted floor markings adding believable context.
OpenAI: GPT-6 Luna: Model B presents the requested objects in a clearer, more literal arrangement, with the box directly on the blue pallet and the hard hat visibly between the hand truck and pallet; it also includes workers and believable dock context. Model A is polished, but the box rests on the flour sacks and the hard hat is partially cropped, making the layout less unambiguous. (Second judge pass, order swapped — scores are the average of both: Model B clearly shows the requested wide loading-dock arrangement | Anthropic: Claude Fable 5.1: Model B satisfies the exact layout — hand truck…
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 requested chair, books, tea, and warm window light while keeping the scene uncluttered and avoiding obvious forbidden objects. Model B presents a fuller nook, but the trees and foliage clearly visible through the window conflict with the strict no-plants constraint. (Second judge pass, order swapped — scores are the average of both: Model B presents the full, inviting reading nook with the requested chair, stacked books, tea, window, and warm afternoon light, while avoiding | Anthropic: Claude Fable 5.1: Both models correctly omit plants, lamps, and ar…
Harbor survey team scene
A coherent multi-subject scene of five people at work on a windy harbor pier during golden-hour light, cinematic 16:9 composition: a marine surveyor kneels over an open orange instrument case in the foreground, an engineer in a teal rain jacket points at a tablet map, a dockworker coils a thick rope beside stacked crab pots, a supervisor in a cream beanie speaks into a radio near a striped bollard, and a drone pilot at the far right monitors a small quadcopter hovering above the water; the group should feel naturally connected in one balanced frame, realistic yet slightly stylized editorial photography.
OpenAI: GPT-6 Luna: Model B presents a more coherent, balanced harbor team and includes the key requested activities and props, with a clear golden-hour setting. Model A has appealing light and detail, but its awkwardly cropped panoramic framing and less clearly depicted rope-coiling role make it less faithful overall. (Second judge pass, order swapped — scores are the average of both: Model B presents a coherent five-person harbor team with the orange instrument case, tablet map, coiled rope and crab pots, radio, | Anthropic: Claude Fable 5.1: Model B delivers all five specified subjects cor…
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 uses the requested colors and includes mountains and a campsite, but the oversized, heavily cropped tent and soft shading undermine the balanced flat-vector look. Model B presents a clear, polished campsite scene with a cohesive four-color palette and strong composition. (Second judge pass, order swapped — scores are the average of both: Model B presents a clear, balanced mountain campsite at dusk in crisp flat-vector shapes and the requested four-color palette. Model A uses the palette | Anthropic: Claude Fable 5.1: Model A violates the core constraint with an obv…
Color-bound workshop still life
A meticulously arranged product-photography still life on a metal workbench in a small watchmaker’s studio, shot straight-on with crisp softbox lighting and shallow depth of field: a matte cobalt-blue ceramic mug, a brushed copper caliper, a translucent amber glass vial, a forest-green velvet ring box, and a chalk-white plastic loupe, each object clearly separated with no color spill or swapping of materials, ultra-realistic commercial photography, 16:9.
OpenAI: GPT-6 Luna: Model B presents all five requested objects clearly separated in a balanced, watchmaker-studio composition, with the specified colors and materials readily recognizable. Model A has convincing workshop detail, but the oversized, partially cropped objects and glossy mug weaken the product-photography arrangement. (Second judge pass, order swapped — scores are the average of both: Model B presents all five requested objects clearly separated on a metal workbench, with a convincing watchmaker’s stu | Anthropic: Claude Fable 5.1: Model B nails more of the color-and-material co…
Readable union hall poster
A 1980s retro print style poster pinned on a corkboard in a busy municipal break room, lit by warm fluorescent overheads, centered and easy to read, 16:9. The poster features bold, legible typography that clearly says: "NIGHT SHIFT / TUESDAY 11:40 / BAY 7" with each line spelled exactly, high contrast ink, simple geometric accents, and no extra words competing with the main text.
OpenAI: GPT-6 Luna: Model B presents a polished retro-style poster centered on the corkboard, with the requested wording clearly readable and the busy, warmly lit break room providing context. Model A has strong legibility, but the poster is oversized and off-center, and its text is arranged across the lines less faithfully. (Second judge pass, order swapped — scores are the average of both: Model B presents the exact requested wording in a centered, high-contrast retro poster with clear geometric accents and a con | Anthropic: Claude Fable 5.1: Model B delivers a genuinely 1980s retro poster…
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 clearly shows the teapot, glass of water, red apple, and window, with convincing reflections across the chrome and polished counter. Model A has dramatic close-up lighting, but the composition is hard to read, the apple is marked with stray text, and the required reflections are less clearly represented. (Second judge pass, order swapped — scores are the average of both: Model B presents the requested teapot, water glass, apple, dark marble, and window reflections in a cohesive photoreal | Anthropic: Claude Fable 5.1: Model A delivers dramatic side light and a bold…
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 follows the ukiyo-e tradition, with a dramatic stylized cresting wave, traditional fishing boats, and a muted dawn palette. Model A has attractive wave detailing, but its modern-looking vessel and more gradient-heavy rendering feel less faithful to the requested woodblock style. (Second judge pass, order swapped — scores are the average of both: Model B strongly matches the ukiyo-e tradition with flat indigo-and-cream planes, bold outlines, stylized foam, and a traditio | Anthropic: Claude Fable 5.1: Model B is unmistakably ukiyo-e: Hokusai-style…
Matchup powered by OpenRouter.