Head to head: Wan vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
Wan vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
By Ryan Merket · Published
Eight image prompts put Wan and Google’s Nano Banana 2 through tests of composition, visual style, scene detail and precise instruction-following. The comparison looks at where each model is dependable—and where a vivid image can still miss the brief.
This is a close matchup, not a coronation. Nano Banana 2’s aggregate score is 64.3 to Wan’s 59.8, but the result carries just limited confidence that either model is genuinely better. Across the eight tasks, Google takes four, Wan takes three, and the cliff-railway poster is a tie: the practical verdict is that they’re effectively even here. Google’s edge comes from making busy scenes feel complete. Its monsoon portrait conveys exhausted joy more clearly, its ukiyo-e wave includes the fishing boat Wan’s image obscures, and its neon harbor supplies the boats, water and luggage-laden scooter the prompt calls for. It also makes the restricted-palette campsite more immersive, though neither model perfectly maintains the strict no-gradients constraint. Wan is stronger when restraint and spatial discipline matter. Its caravanserai keeps the courtyard symmetrical and free of apparent vehicles; its corridor has a cleaner, uninterrupted vanishing-point composition. In the reflections task, Wan’s arrangement is cleaner, while Google’s more obvious reflections come with an unrequested hand. Those are different kinds of prompt fidelity, not a decisive gap in capability. The poster result captures the matchup: Google’s version gets the requested details on the page, while Wan’s is a coherent, crisp design, and neither earns a consistent advantage across the judging. **Final call: tie. Nano Banana 2 leads on the raw score, but the confidence and task-by-task split do not support declaring it the better model.**
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 makes the window and red apple reflections more evident across the chrome and water, though the hand is an unrequested distraction. Model A has a cleaner arrangement, but the required reflections in the water are less clear. (Second judge pass, order swapped — scores are the average of both: Model B better conveys the dramatic side light and shows the window and apple reflected in the chrome, with the apple also visible through and near the water’s surface. Model A is cleaner and well co | Anthropic: Claude Fable 5.1: Model A delivers the requested clean close-up w…
Monsoon relief portrait
A cinematic travel photograph of a solo ferry captain on the windblown dock at Bhairava Inlet just as a week-long monsoon finally breaks, his face and body language convincingly showing unmistakable relief mixed with exhausted joy, rainwater still dripping from his wool cap, one hand clutching a coil of blue rope, gulls circling behind him, weathered teal boat hull at frame left, shallow depth of field, realistic skin detail, dramatic overcast light splitting into a first warm sunbeam, medium close-up composition, 16:9
OpenAI: GPT-6 Luna: Model B conveys the requested exhausted joy and relief more unmistakably, while also including the blue rope, teal ferry, circling gulls, rain-darkened dock, and warm break in the clouds. Model A is a strong, clean cinematic image, but its expression reads as a more restrained smile and less clearly shows the emotional release. (Second judge pass, order swapped — scores are the average of both: Model B better conveys unmistakable exhausted joy, with a vivid cap-touching gesture and a cinematic s | Anthropic: Claude Fable 5.1: Model B nails the emotional brief with a convin…
Cliff railway poster text
A crisp vintage-inspired travel poster for the fictional coastal funicular at Veyra Cliffs, rendered in vibrant flat-vector design, showing terraced sea cliffs, a small red rail car, stylized whitecaps, and a clean central poster panel with large fully legible typography that reads exactly: "VEYRA CLIFF RAIL" on the first line and "LAST CAR 18:40" on the second line, high contrast lettering, elegant layout, print-ready clarity, balanced composition, 16:9
OpenAI: GPT-6 Luna: Model B more fully delivers the terraced coastal cliffs, red funicular, whitecaps, and a clean central panel with the exact, legible wording. Model A is crisp and attractive, but its typography sits at the top rather than in a central panel and its rail car feels less integrated into the scene. (Second judge pass, order swapped — scores are the average of both: Model B depicts the terraced cliffs, red rail car, whitecaps, and exact wording, but its nested framing and duplicated scenery make the | Anthropic: Claude Fable 5.1: Model A delivers a coherent single 16:9 flat-vec…
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 B has a clearer, more immersive campsite composition and stronger visual polish, while both entries largely honor the four-color flat-vector brief. Model A includes the requested camping details but feels more crowded and has some awkwardly rendered objects; both show slight tonal variation that softens the strict no-gradients constraint. (Second judge pass, order swapped — scores are the average of both: Model B delivers a richly composed, clearly recognizable dusk campsite with a crescen | Anthropic: Claude Fable 5.1: Model A keeps a strict four-color flat palette…
Empty caravanserai square
A wide-angle architectural photograph of the ancient caravanserai square at Orun Dast during pale morning light, sandy stone arcades, patterned blue tile fountain, two travelers walking with satchels, a few resting donkeys near the shade, long quiet shadows, documentary realism, symmetrical composition, and absolutely no vehicles of any kind anywhere in the scene — no cars, buses, bicycles, carts, scooters, or wagons, 16:9
OpenAI: GPT-6 Luna: Model A closely matches the requested quiet, symmetrical courtyard with blue-tiled fountain, two travelers, resting donkeys, and no visible vehicles. Model B is more elaborate and visually rich, but appears to include small wheeled objects in the background, conflicting with the explicit vehicle-free requirement. (Second judge pass, order swapped — scores are the average of both: Model B has a more monumental, richly detailed composition, but appears to include a parked vehicle near the far arca | Anthropic: Claude Fable 5.1: Model A delivers a clean symmetrical caravanser…
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 completely depicts the fishing boat riding the cresting wave at dawn, while retaining the ukiyo-e flat colors, bold outlines, and stylized foam. Model A has a strong woodblock-print look and appealing palette, but the boat is missing or obscured, weakening its match to the prompt. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows a fishing boat riding the cresting wave at dawn, while retaining the bold outlines, stylized foam, and muted i | Anthropic: Claude Fable 5.1: Both deliver convincing Hokusai-style flat colo…
Neon harbor midnight
An ultra-detailed digital painting of a narrow harbor lane in Port Seln at neon-soaked midnight, wet cobblestones reflecting magenta noodle-shop signs and cyan hotel lights, a sleepy luggage scooter parked beside stacked crab traps, distant black water between buildings, faint mist in the air, the lighting unmistakably deep night with vivid neon glow shaping the entire mood, cinematic perspective from street level, rich reflections, 16:9
OpenAI: GPT-6 Luna: Model A captures the vivid neon, wet cobbles, mist, and crab traps, but the harbor water and luggage on the scooter are not clearly shown. Model B includes a more convincing harbor lane with boats and distant water, plus a luggage-laden scooter and strong neon reflections, while maintaining a richly detailed midnight atmosphere. (Second judge pass, order swapped — scores are the average of both: Model B more fully realizes the harbor setting, with a luggage-laden scooter beside crab traps, vivid | Anthropic: Claude Fable 5.1: Model B delivers a richly detailed, atmospheric…
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 more cleanly fulfills the prompt with a long, empty, symmetrical aisle converging at one clear vanishing point and consistent scale recession. Model B is attractive and strongly perspectival, but its repeated architectural bays make the aisle feel less like one uninterrupted corridor. (Second judge pass, order swapped — scores are the average of both: Model B combines a clear centered vanishing point and convincing scale recession with the requested warm overhead lighting, while Model A | Anthropic: Claude Fable 5.1: Model A delivers a clean, geometrically consiste…
Matchup powered by OpenRouter.