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
Wan and Google’s Nano Banana 2 face off across eight image-generation tasks, from exact object counts and constrained palettes to cinematic scenes and poster text. The comparison tests both prompt fidelity and visual execution across very different briefs.
Google’s Nano Banana 2 takes the saved verdict, scoring **65.9** to Wan’s **57.5**. It won six tasks to Wan’s one, with one tie—a decisive result on this task sample, not a statistical claim. Nano Banana 2’s strongest showing came on visually demanding scenes: its clockwork moth had finer macro detail, its star-ark more convincingly dwarfed the harbor, and its cyberpunk bazaar delivered richer atmosphere and more legible signage. It also edged the named-style and moonport-poster tasks, including the exact, readable poster text. Wan had a meaningful win on exact counting: its seven espresso-filled cups were clearer and more faithful than Nano Banana 2’s cups, which appeared empty and came with unrequested saucers. The restricted-palette campsite was a tie; Wan adhered more tightly to the four-color constraint, while Google offered a more polished, textured composition. **Verdict: Nano Banana 2 wins this matchup decisively on the saved task sample, while Wan shows an edge where literal constraint-following matters most. This assessment applies only to the saved task sample.**
Exact counting
A clean overhead flat lay on white marble of EXACTLY seven identical espresso cups arranged in a single straight row, each clearly separate and individually countable, even soft lighting, 16:9.
OpenAI: GPT-6 Luna: Model A shows exactly seven clearly countable espresso-filled cups in a straight row, closely matching the requested objects. Model B also has seven cups in a row, but adds saucers and the cups appear empty, making it less faithful to the prompt. (Second judge pass, order swapped — scores are the average of both: Both images show exactly seven cups in a straight row on white marble, but Model A makes the espresso cups immediately recognizable with visible espresso and clear separation. Model B i | Anthropic: Claude Fable 5.1: Both renders deliver exactly seven countable es…
Clockwork moth detail
An extreme close-up macro illustration of a clockwork moth resting on black velvet in a wizard-engineer’s workshop, rendered in hyper-detailed dark fantasy realism: each brass scale on the wings etched with tiny runes, frayed velvet fibers individually visible, hair-thin antennae, minute screws, pollen dust, and faint oxidation on silver joints, all sharply resolved without smudging; dramatic raking lamplight from the left, shallow background falloff, centered composition.
OpenAI: GPT-6 Luna: Model B better delivers the extreme macro view, intricate clockwork construction, sharply visible velvet fibers, and dark workshop atmosphere. Model A has a more symmetrical, centered silhouette and clearer spread wings, but feels less close-up and its rune-like markings are less convincing as fine etching. (Second judge pass, order swapped — scores are the average of both: Model B better delivers the extreme macro realism, with exceptionally visible velvet fibers, intricate clockwork, and drama | Anthropic: Claude Fable 5.1: Model A offers a clean, centered, clearly left-…
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 more clearly shows the blue cylinder behind both objects and gives the green cube a convincing matte finish, while preserving the requested left-right arrangement and duck placement. Model A is also polished and faithful, but its cylinder is more heavily obscured and the cube looks less matte. (Second judge pass, order swapped — scores are the average of both: Model B more clearly renders a crisp matte cube and presents the objects in a balanced studio composition; Model A also follows t | Anthropic: Claude Fable 5.1: Both bind every color to the correct object wit…
Colossus harbor perspective
A wide cinematic matte-painting view of the stormbreak harbor on the planet Ilyra, where a colossal docked star-ark rises behind a row of ordinary fishing skiffs and warehouse cranes; correct scale is crucial—the skiffs should look tiny, workers on the pier nearly specks, and the ark’s lower hull should dwarf the buildings—while all pier planks, rails, and rooflines recede consistently toward a single vanishing point; cool dawn light, sea mist, realistic atmospheric perspective, 16:9.
OpenAI: GPT-6 Luna: Model B more convincingly depicts a colossal star-ark dwarfing the harbor, with cinematic dawn mist and a richer, more futuristic setting. Model A has clear skiffs and pier details, but its ark reads more like a conventional ship and the scale contrast is less dramatic. (Second judge pass, order swapped — scores are the average of both: Model B more clearly depicts a colossal star-ark dwarfing the harbor, with tiny skiffs, atmospheric dawn mist, and a richly cinematic matte-painting feel. Model | Anthropic: Claude Fable 5.1: Model B delivers a true colossal star-ark with a…
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 captures the requested ukiyo-e style, dawn, and muted indigo-and-cream palette, though the fishing boat is partly obscured. Model B has a more fully realized woodblock composition with a dramatic cresting wave and clearly depicted fishermen, while its extra boats and lettering add some visual clutter. (Second judge pass, order swapped — scores are the average of both: Model B more fully evokes ukiyo-e through its intricate stylized foam, bold outlines, indigo-and-cream palette, and detai | Anthropic: Claude Fable 5.1: Model B delivers a richer, more convincing ukiy…
Dragon market without people
At dusk in a narrow cyberpunk bazaar alley on the floating city of Vey Orin, glowing magenta and teal lanterns reflect on rain-slick cobblestones while a tethered pack-dragon loaded with sealed spice crates stands beside a noodle stall and humming power cables crisscross overhead; the scene must contain no people at all—no vendors, no shoppers, no silhouettes, no reflections of anyone—cinematic neon lighting, richly atmospheric, eye-level wide shot, 16:9.
OpenAI: GPT-6 Luna: Model A more clearly shows a tethered, upright pack-dragon, while Model B creates a richer cyberpunk bazaar with stronger depth, more legible Vey Orin and noodle signage, and vivid neon reflections. Both keep the scene people-free, but Model B’s atmosphere and composition give it the edge despite the dragon being less clearly tethered. (Second judge pass, order swapped — scores are the average of both: Model B more fully realizes the cyberpunk bazaar with the Vey Orin signage, noodle stall, dens | Anthropic: Claude Fable 5.1: Model B delivers a convincingly cyberpunk bazaa…
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 adheres more closely to the strict four-color, flat-vector constraint and clearly depicts a campsite at dusk, though its many small props make the composition busy. Model B has a more polished, expansive scene, but its extra tonal variation and textured appearance weaken adherence to the restricted palette. (Second judge pass, order swapped — scores are the average of both: Model B has a more balanced, immersive campsite composition and stronger visual polish, with the dusk setting clear | Anthropic: Claude Fable 5.1: Model A strictly honors the four-color constrai…
Moonport poster text
A retro-futurist travel poster pinned inside a lunar ferry terminal, designed in bold mid-century screenprint style with crisp vector lettering and subtle paper grain, showing a sapphire-ringed gas giant above a silver launch pad; the poster must prominently and legibly read exactly: "MOONPORT KHEPRI 7" on the top line, "Departures 04:35" beneath it, and "Dock C" in a small boxed label at the lower right; clean layout, high contrast inks, straight-on composition, even gallery lighting, 16:9.
OpenAI: GPT-6 Luna: Model B makes the silver lunar launch pad and sapphire-ringed gas giant unmistakable, while keeping all three requested text elements exact and legible. Model A has a cleaner, more minimal mid-century layout, but its launch-pad imagery is less clear and the overall poster is less visually engaging. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested wide mid-century screenprint poster, with the exact text clearly legible, a vivid ringed planet | Anthropic: Claude Fable 5.1: Model A delivers a clean, high-contrast mid-cen…
Matchup powered by OpenRouter.