Head to head: Meta Muse Image Edit vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
Meta Muse Image Edit vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
By Ryan Merket · Published
A head-to-head across eight image tasks tests detailed scene composition, precise constraints, and subject clarity—from handwork and action scenes to exact counts and restricted palettes.
Meta Muse Image Edit leads the saved results with an aggregate score of **65.5**, against **59.1** for Google: Nano Banana 2 (Gemini 3.1 Flash Image). The task record is five wins for Meta, two for Google, and one tie. This is a clear result on this sample, not a claim about performance beyond it. Meta’s edge came from stronger prompt-specific composition: it won on the bound-gear color trial, mantis-glove macro, canal-rooftop parkour relay, negation, and rain-visor reflection. Its images more clearly surfaced the requested material details, subjects, actions, or constraints—especially the layered visor reflections and the relay’s six distinct roles. Google won the restricted-palette and exact-count tasks. Its campsite scene was judged more balanced and immersive despite both images straying from the strict four-color rule; its row of seven espresso-filled cups also matched the request more specifically. Hands and anatomy was a tie: Meta made the grip and threadwork clearer, while Google gave the bracelet-making scene a stronger candid setting. **Verdict: Meta Muse Image Edit wins this saved task sample, 65.5 to 59.1.** The advantage rests on more frequent, more faithful execution of the requested visual priorities—not a guarantee of broader superiority.
Hands & anatomy
A natural candid photo of a person tying a friendship bracelet, both hands clearly visible with anatomically correct fingers and a plausible grip on the threads, soft daylight, shallow depth of field, 16:9.
OpenAI: GPT-6 Luna: Model A makes both hands and the thread manipulation the clear focus, with convincing anatomy, a plausible grip, soft daylight, and shallow depth of field. Model B has a stronger candid setting and clearly shows the person making a bracelet, but the hands and action are less prominent in the composition. (Second judge pass, order swapped — scores are the average of both: Model A gives a clearer, more detailed view of both hands and their plausible grip on the threads, with soft daylight and shal | Anthropic: Claude Fable 5.1: Model A delivers large, sharply rendered hands…
Bound Gear Color Trial
A hyper-clean studio product arrangement of five training objects on a neutral graphite backdrop, each with its own exact color and material and no color swapping: a brushed titanium kettlebell with a saffron-yellow handle, a translucent emerald-green water bottle with a black cap, a matte crimson medicine ball with charcoal stitching, a glossy cobalt-blue fencing mask with a silver mesh face, and a pale ash-wood skateboard deck with violet wheels; soft directional light, subtle shadows, premium commercial realism, objects clearly separated so the color-material bindings are unambiguous, 16:9.
OpenAI: GPT-6 Luna: Model A clearly presents all five requested objects with the specified color-material pairings, clean separation, and polished studio lighting. Model B includes the set but the medicine ball reads as a basketball, the skateboard is cropped, and the arrangement has less balanced spacing. (Second judge pass, order swapped — scores are the average of both: Model A renders all five objects with especially clear material and color details, including the medicine ball’s charcoal stitching, and keeps t | Anthropic: Claude Fable 5.1: Model A renders all five objects with correct c…
Mantis Glove Macro
An extreme macro close-up of a rock climber’s chalked right hand pinching a tiny brass bolt-on hold shaped like a praying mantis on a speckled overhang wall, photographed with a true macro lens look: razor-thin depth of field, the textured skin, chalk dust, and engraved serial number on the hold tack sharp while the background route markers dissolve into creamy bokeh; cinematic sports advertising style, cool warehouse light with a narrow warm rim light, diagonal composition, no extra hands in frame.
OpenAI: GPT-6 Luna: Model A more fully delivers the extreme macro look, chalky skin detail, dramatic cool-and-warm lighting, and creamy route-marker bokeh, with a striking diagonal composition. Model B has a clearly legible hold inscription, but its wider framing and less distinctive lighting feel less faithful to the requested macro advertising image. (Second judge pass, order swapped — scores are the average of both: Model A makes the praying-mantis shape especially unmistakable and delivers a more dramatic macro | Anthropic: Claude Fable 5.1: Model A delivers the requested extreme-macro lo…
Canal Rooftop Parkour Relay
A dynamic multi-subject cyberpunk action illustration of six parkour athletes mid-relay on linked rooftops above a narrow canal market at dawn: one athlete in a white windbreaker vaults a gap, a second in burnt orange crouches to pass a glowing baton backward, a third in lime sleeves lands on a tilted solar panel, a fourth swings from a laundry line, a fifth spots the route from a rust-red water tower, and a sixth merchant on a balcony reacts with surprise below; the scene must feel coherent and well-balanced, with believable spacing, clear interactions, layered depth, neon shop signs reflected in the canal, misty backlight, and a strong left-to-right compositional flow, 16:9.
OpenAI: GPT-6 Luna: Model A more fully realizes the prompt with all six roles, a vivid neon canal market, misty dawn depth, and a clear left-to-right relay flow. Model B depicts the requested actions clearly, but feels more schematic and less immersive, with less convincing environmental depth and market detail. (Second judge pass, order swapped — scores are the average of both: Model A more clearly conveys the relay, including the backward baton pass, the laundry-line swing, and the route spotter atop the water to | Anthropic: Claude Fable 5.1: Model A delivers all six subjects with clear ro…
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 depicts the requested cozy chair, stacked books, steaming tea, and warm afternoon light while keeping plants, lamps, and artwork out of the frame. Model B is appealing and well composed, but visible outdoor foliage through the window conflicts with the strict no-plants requirement. (Second judge pass, order swapped — scores are the average of both: Model B has a lovely, balanced nook, but visible foliage outside the window conflicts with the instruction to have no plants anywhere in fram | Anthropic: Claude Fable 5.1: Model A satisfies every constraint—armchair, bo…
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 more balanced, polished campsite scene with stronger depth and clearer focal details. Both images stray from the exact four-color restriction, but Model A also feels more crowded and less cohesive. (Second judge pass, order swapped — scores are the average of both: Model B has richer campsite detail and a more immersive composition, but its sky gradient and many color variations conflict with the strict four-color, no-gradient requirement. Model A is more consistently flat and pale | Anthropic: Claude Fable 5.1: Model A stays a bit closer to a strict flat fou…
Rain Visor Reflection Sprint
A night-time track cycling pit scene in a realistic editorial sports photo style: a female keirin rider in a matte cobalt skinsuit stands beside her bike under magenta-and-teal stadium signage, wearing a crystal-clear rain visor over her helmet and mirrored chrome shoe covers; her visor must accurately show the reflected oval track, a mechanic in an orange jacket, and streaks of LED sponsor lights, while raindrops on the visor and the puddled concrete refract and distort those same elements correctly; include a glass timing booth behind her with the interior visible through the panels and faint reflections layered on top, dramatic wet-surface lighting, crisp detail, 16:9.
OpenAI: GPT-6 Luna: Model A more directly fulfills the distinctive visor-reflection requirement, with the orange mechanic, track curve, and LED streaks visibly layered across the rain-speckled visor; it also delivers atmospheric wet-surface lighting and a visible glass booth interior. Model B shows the full rider, bike, chrome shoe covers, and pit scene clearly, but its visor lacks the requested reflected details and the frame feels busier. (Second judge pass, order swapped — scores are the average of both: Model A | Anthropic: Claude Fable 5.1: Model A makes the visor reflections the hero (t…
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 separated cups in a straight row, while Model B also meets the exact count and adds visible espresso, making the subject more specific to the prompt. Model B has a polished, balanced flat-lay presentation, though both images are clean and technically strong. (Second judge pass, order swapped — scores are the average of both: Model B shows exactly seven clearly separate espresso-filled cups in a straight overhead row, matching the espresso-cup request more dire | Anthropic: Claude Fable 5.1: Both Model A and Model B deliver exactly seven…
Matchup powered by OpenRouter.