Head to head: Google: Nano Banana 2 (Gemini 3.1 Flash Image) vs Wan
Google: Nano Banana 2 (Gemini 3.1 Flash Image) vs Wan
By Ryan Merket · Published
Eight image-generation prompts tested object relationships, text rendering, art direction, palette control, and detailed scene adherence. The comparison rewards both precise, restrained compositions and images that bring complex briefs to life.
Google’s Nano Banana 2 takes the matchup, scoring 67.2 to Wan’s 56.9 and winning six of eight tasks. The result is backed by limited confidence—a clear verdict, not a photo finish. Its advantage came in the briefs with the most to juggle. Google made the heron’s basket repair legible, packed the musk ox archivist’s workshop with requested artifacts, and delivered the stronger Art Deco pangolin poster. It also handled the restricted-palette campsite, ukiyo-e fishing scene, and rain-streaked kingfisher hatchery more convincingly. Wan had two distinct wins: attribute binding and legible multi-line text. Its clean studio arrangement made the colored objects and their relationships easier to read, and its three lines of text were crisp, correct, and better balanced. Those are meaningful strengths, especially when precision and uncluttered presentation matter most. But across this set, Google was more reliable when a prompt demanded several specific details within a cohesive image. **The call: Nano Banana 2 wins, 6–2, on stronger overall execution of complex visual briefs; Wan’s edge is in clean, controlled layouts.**
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: Both images correctly show the green cube to the left of the glossy red sphere, with the blue cylinder behind and the yellow duck on top of the cube. Model B presents the objects more clearly in a clean studio composition, while Model A’s rustic background is more distracting. (Second judge pass, order swapped — scores are the average of both: Model B cleanly matches the requested object placement, colors, materials, and soft studio tabletop presentation. Model A also gets the object relationshi | Anthropic: Claude Fable 5.1: Both correctly bind all colors and place the gr…
Heron reed-basket repair
A naturalistic wildlife documentary still of a field biologist kneeling on a cedar boardwalk at dawn, carefully repairing a storm-damaged reed basket nest while a black-crowned night heron watches from arm’s length; her face, elbows, knees, and both bare hands are fully visible, with all ten fingers clearly separated and naturally posed as she threads pale sedge through the weave, realistic anatomy and proportion, damp waders, fog lifting off the marsh, dew on cattails, long-lens compression, soft backlight with cool mist and warm sunrise rim, cinematic 16:9.
OpenAI: GPT-6 Luna: Model A more clearly conveys a damaged basket being repaired, with the heron prominently perched at arm’s length and a convincing misty dawn boardwalk scene. Model B depicts the black-crowned night heron more recognizably, but its basket looks comparatively intact and the repair action is less apparent. (Second judge pass, order swapped — scores are the average of both: Model A more clearly shows a storm-damaged basket being repaired amid dawn mist, with a strong documentary composition and natu | Anthropic: Claude Fable 5.1: Model A delivers a convincing documentary frame…
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 B presents all three lines crisply and correctly in a clean, centered hierarchy on a subtle-grain navy background. Model A also renders the text accurately, but the inset card and large unused upper area make it feel less minimal and balanced. (Second judge pass, order swapped — scores are the average of both: Model B presents all three lines crisply, correctly, and in a balanced centered layout that closely matches the prompt. Model A renders the text accurately too, but the large empty u | Anthropic: Claude Fable 5.1: Model A renders all three lines perfectly spell…
Musk ox archivist workshop
An ultra-detailed stop-motion claymation scene of a tundra conservator’s workbench inside a remote wildlife station, where a tagged musk ox pelt sample, braided qiviut yarn, frost-crusted binocular straps, lichen-covered antlers, weathered enamel mugs, riveted brass calipers, and a half-disassembled wind meter are arranged in crisp focus; every fiber, weave, scratch, ice crystal, felted fur strand, and oxidized metal edge rendered sharply without smudging, raking side light from a narrow window, deep teal shadows, macro realism, 16:9.
OpenAI: GPT-6 Luna: Model A more completely includes the specified tagged pelt, qiviut yarn, frosted binocular straps, lichen-covered antlers, mugs, calipers, and disassembled wind meter, with crisp detail and atmospheric side lighting. Model B has an appealing wintry workshop but substitutes a musk ox head for the pelt sample and makes several requested objects less explicit. (Second judge pass, order swapped — scores are the average of both: Model A includes nearly all requested artifacts, especially the clearly | Anthropic: Claude Fable 5.1: Model A nails nearly every requested object with…
Art Deco pangolin poster
A faithful Art Deco travel-poster illustration advertising the fictional Mbale Nocturnal Reserve, centered on a ground pangolin ascending a fan-shaped staircase of stylized termite mounds beneath a crescent moon; bold geometric symmetry, stepped forms, sunburst motifs, streamlined silhouettes of acacia trees and hornbills, limited palette of obsidian, jade, parchment, and burnished gold, elegant period lettering integrated into the composition, flat graphic shadows, crisp screen-printed finish, portrait orientation.
OpenAI: GPT-6 Luna: Model A delivers a polished portrait Art Deco travel poster with a clearly recognizable pangolin ascending a dramatic fan-shaped staircase, a crescent moon, wildlife, and legible integrated lettering. Model B includes several requested motifs, but its wide canvas and narrow centered poster weaken the composition, and its pangolin is less clearly shown ascending the staircase. (Second judge pass, order swapped — scores are the average of both: Model A is the more polished Art Deco travel poster, | Anthropic: Claude Fable 5.1: Model A delivers a true portrait Art Deco travel…
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 delivers a cohesive, detailed dusk campsite scene with a balanced layout and a clean four-color flat-vector style. Model B is also graphic and recognizable, but its oversized figure and object-heavy foreground feel less balanced, and some gray tones appear outside the specified palette. (Second judge pass, order swapped — scores are the average of both: Model A has a more balanced, polished campsite scene and reads clearly as a dusk landscape; Model B includes the requested motifs but fe | Anthropic: Claude Fable 5.1: Model A delivers a polished, well-balanced flat…
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 A makes the fishing crew and boat prominent amid a dramatic, traditionally styled cresting wave, with a dawn sky and muted indigo-and-cream palette. Model B is elegant and stylistically faithful, but its boat and figures feel less clearly like a fishing scene, and the extra partial boat at the right is distracting. (Second judge pass, order swapped — scores are the average of both: Model B more directly matches the requested single fishing boat, dawn, muted indigo-and-cream palette, and fl | Anthropic: Claude Fable 5.1: Model A delivers a richly executed ukiyo-e prin…
Kingfisher hatchery glasshouse
A photoreal image of a riverside glass incubation house at blue hour, viewed from just outside a rain-speckled pane: inside, a wildlife rehabilitator in a saffron rain jacket tends three rescued kingfisher chicks beneath amber heat lamps, while the glass shows correct reflections of alder branches, a bicycle leaning on the railing, and the cloudy sky, plus subtle refraction and distortion through curved jars, water beakers, and a chrome kettle; a shallow puddle in the foreground mirrors the lit interior, accurate transparency, reflection, and light behavior, cinematic 16:9.
OpenAI: GPT-6 Luna: Model A more closely captures the rain-speckled pane, reflective blue-hour atmosphere, saffron-clad rehabilitator, warm lamps, and requested glassware while maintaining a cinematic composition. Model B is cleanly composed and includes the riverside setting and three birds, but its glass appears less rain-speckled and the birds read less clearly as rescued chicks being tended. (Second judge pass, order swapped — scores are the average of both: Model A more clearly conveys the rain-speckled pane, | Anthropic: Claude Fable 5.1: Model A delivers a convincingly photoreal blue-h…
Matchup powered by OpenRouter.