Head to head: Google: Nano Banana 2 (Gemini 3.1 Flash Image) vs Meta Muse Image Edit
Google: Nano Banana 2 (Gemini 3.1 Flash Image) vs Meta Muse Image Edit
By Ryan Merket · Published
A close-scoring image-model matchup tests more than polish: precise object placement, perspective, lighting and material rendering all matter. The results favor the model that more often gets the whole brief right, while leaving room for a couple of standout wins on the other side.
Meta Muse Image Edit takes the matchup, 65.8 to 63.3, with an limited confidence verdict and five task wins to Google’s two, plus one tie. That’s a clear result across the test set—not a blowout, given the narrow aggregate score gap. Muse’s edge comes from breadth. It wins the midnight repair, showroom, spatial-layout, charging-plaza and lantern-workbench tasks, where the verdicts favor its handling of scene relationships, requested details and convincing presentation. The desk-under-window placement and the workbench’s carefully arranged objects are especially telling: attractive images matter, but so does putting the right things in the right places. Nano Banana 2 has its own concrete strengths. It wins on perspective and scale, producing the more expansive library aisle with stronger depth cues, and on the restricted-palette scene, where it adheres more closely to the color constraint. Attribute binding is a tie: both models place the colored objects correctly, with neither earning a decisive overall advantage. The takeaway is not that Muse dominates every kind of image task. It is that Muse is more consistently dependable across varied briefs, while Nano Banana 2’s wins are concentrated in specific visual strengths. **Verdict: Meta Muse Image Edit wins on consistency, not by a landslide.**
Midnight repair realism
Ultra-photorealistic night-time apartment scene, 16:9: a robotics engineer repairing a compact home server on a walnut dining table lit by a single warm desk lamp and faint blue city glow from a nearby window; brushed aluminum panels, braided black cables, matte silicone tool handles, woven gray hoodie fabric, natural skin texture, and the varnished wood grain must all look physically accurate; shadows should fall convincingly across scattered Torx screws and a ceramic mug, with subtle specular highlights on metal edges and realistic light falloff in the room. Cinematic composition, shallow depth of field, believable real-world materials.
OpenAI: GPT-6 Luna: Model B more tightly emphasizes the repair and gives the aluminum, braided cables, hoodie, and walnut table a convincing cinematic look with the requested warm-and-blue lighting. Model A also matches the scene well, but its wider framing and prominent branded mug make the repair feel slightly less focused. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the prompt, with convincing warm lamp light, blue city glow, physically detailed server compone | Anthropic: Claude Fable 5.1: Model B delivers the stronger midnight-repair r…
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 has a particularly clear, centered vanishing point and convincing, consistent reduction in book and shelf scale, with warm lighting and a richly rendered library interior. Model B also follows the one-point perspective well, but its aisle feels a little less expansive and its distant endpoint is less visually compelling. (Second judge pass, order swapped — scores are the average of both: Model A presents a longer, more clearly converging aisle with consistent depth cues and warm lighting | Anthropic: Claude Fable 5.1: Model B delivers a rigorously symmetric one-poi…
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 place the yellow duck on the green cube and keep the blue cylinder behind the cube and red sphere, but Model A better matches the requested 16:9 tabletop scene and gives the cube a more clearly matte finish. Model B is polished and clean, though its framing is less faithful to the specified aspect ratio and its objects feel oversized. (Second judge pass, order swapped — scores are the average of both: Model B presents all four objects in the requested positions with clear color and m | Anthropic: Claude Fable 5.1: Both outputs bind every color to the correct ob…
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 presents a cleaner, more polished flat-vector campsite with a clear dusk cue and stronger use of the requested four-color palette. Model A has appealing layered scenery, but the textured outer border introduces off-palette variation and the campsite feels crowded near the bottom. (Second judge pass, order swapped — scores are the average of both: Model B more closely follows the flat-vector, 16:9 brief and keeps the scene largely to the four requested colors, while Model A adds a texture | Anthropic: Claude Fable 5.1: Model A keeps strictly to the four flat colors,…
Showroom glass and chrome
High-end photorealistic electronics showroom, 16:9: a smoked-glass display case contains a pearl-white smart helmet and a chrome foldable e-bike headlight module; in front of the case stands a woman in a moss-green coat holding a coral phone, while behind her a wall of tiny amber indicator LEDs glows. The glass case must show correct reflections of the woman and the LED wall, the smoked glass must partially reveal the objects inside with realistic transparency, and the chrome headlight module must mirror its surroundings with accurate distorted reflections; include a polished concrete floor with a faint reflected streak from the LEDs. Clean commercial photography lighting, precise optics, realistic refraction and glare.
OpenAI: GPT-6 Luna: Model A more closely matches the requested 16:9 showroom scene, with the helmet and headlight module clearly visible inside a partially transparent case and the woman and amber LEDs reflected in the glass. Model B has striking chrome reflections and polished lighting, but its framing is wider than requested and the chrome module reads more like an oversized sculptural assembly. (Second judge pass, order swapped — scores are the average of both: Model B more clearly places the woman in front of t | Anthropic: Claude Fable 5.1: Model A delivers a convincingly natural showroo…
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 places the desk directly beneath the back-wall window and otherwise matches the requested furniture layout; Model A is more polished and convincingly isometric, but its desk is off to the right of the window. Model B is less dimensional, yet its clear flat-vector look and stronger spatial adherence make it the better fit overall. (Second judge pass, order swapped — scores are the average of both: Model B places all four requested items in the specified locations, especially the desk dire | Anthropic: Claude Fable 5.1: Model B places every element as specified—bed o…
Car-free charging plaza
Pastel storybook illustration, wide 16:9 composition: an early-morning public charging plaza for e-bikes and delivery robots outside the fictional 'LumaLoop Microgrid Library,' with mint-green charging posts, neatly parked scooters, a courier robot waiting beside a map kiosk, planters of lavender, and a barista serving coffee from a tiny solar kiosk. The plaza must be completely free of cars, vans, buses, motorcycles, and any other full-size road vehicles—only the small devices and people in the scene. Soft sunrise light, airy whimsical detail, gentle long shadows, inviting civic-technology atmosphere.
OpenAI: GPT-6 Luna: Model A includes the requested neatly parked scooters alongside e-bikes, charging posts, a courier robot, map kiosk, lavender planters, and a barista, with no full-size road vehicles. Model B is polished and atmospheric, but emphasizes bicycles rather than scooters and has a slightly less lively plaza scene. (Second judge pass, order swapped — scores are the average of both: Model B more closely captures the requested soft sunrise storybook atmosphere, with the library, mint charging posts, lave | Anthropic: Claude Fable 5.1: Model A packs in every requested element (scoot…
Lantern workbench layout
Photorealistic product-workshop scene, 16:9: on a scratched maple workbench, a teal camping lantern sits exactly in the center; a folded copper drone frame is to the left of the lantern; a yellow multimeter is to the right of the lantern; a tiny pile of three silver watch batteries is directly in front of the lantern; a black soldering iron stand is behind the lantern; a transparent zip bag of purple cable ties is between the multimeter and the lantern; and a red microfiber cloth is draped on top of the drone frame. Shot from a slightly elevated three-quarter angle with crisp studio detail and soft overhead daylight.
OpenAI: GPT-6 Luna: Model B presents the requested objects in a clear, convincing arrangement, with the cloth on the copper frame, the purple ties between the lantern and multimeter, and the batteries in front; its crisp shallow-focus workshop lighting is especially effective. Model A also follows the layout closely, but its busier background and less refined visual presentation make it slightly less compelling. (Second judge pass, order swapped — scores are the average of both: Model B follows the requested layout | Anthropic: Claude Fable 5.1: Model B satisfies more of the spatial relations…
Matchup powered by OpenRouter.