Head to head: Flux 3 Image vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
Flux 3 Image vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
By Ryan Merket · Published
A broad image-generation face-off across constraint following, product scenes, action, and exact-count tasks. The results reveal different strengths rather than a decisive gap.
The headline scores are close: Flux 3 Image posts 61.2, while Nano Banana 2 scores 65.0. That 3.8-point edge is not a verdict—the evaluation puts confidence that either model is genuinely better at just 64%, which makes this a sample tie. The task results split, too. Flux gets the nod for negation, keeping plants out of a cozy reading scene, and for the seven espresso-filled cups. Google is credited for the mid-jump courier, the strict four-color campsite, the rainy café checkout, and the eleven lip balms. Both models hit the requested balm count; their difference was presentation, not arithmetic. The golden-hour kettle ad is a tie: Flux better captured the warm light and long shadows, while Google drifted toward blue-hour ambience and a busier frame. Reflections and glass is an especially messy comparison: Flux is listed as the task winner, but the written assessments disagree about which image made the reflections more convincing. That ambiguity fits the wider result—neither model owns every kind of prompt. **Final call: Too close to call.** Google has a narrow aggregate lead and more task wins, but the confidence is too low to crown it; Flux has clear wins of its own, including constraint following and exact counting. For this matchup, they are effectively even.
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 closely follows the constraints, showing the cozy chair, books, tea, and warm window light without visible plants, lamps, or artwork. Model B is also inviting, but the leafy greenery visible outside the window conflicts with the instruction to have no plants anywhere in frame. (Second judge pass, order swapped — scores are the average of both: Model B has a cozy, polished arrangement, but visible foliage through the window violates the explicit no-plants constraint. Model A includes the | Anthropic: Claude Fable 5.1: Model A satisfies all elements with clearly warm…
Golden-hour kettle ad
A photorealistic lifestyle advertising shot of a matte sage-green electric gooseneck kettle on a tiny apartment balcony table, steam curling from the spout beside a half-open paperback and a sliced blood orange, captured at golden hour with long honey-colored shadows, warm rim light on brushed metal, and a calm end-of-day mood; three-quarter composition, shallow depth of field, 16:9.
OpenAI: GPT-6 Luna: Model A more closely delivers the requested honey-golden hour, long shadows, and calm balcony-ad mood, with the kettle, curling steam, open paperback, and sliced blood orange clearly arranged as the focus. Model B includes the key props but feels more like blue-hour city ambience, and its extra objects and prominent book text make the composition busier. (Second judge pass, order swapped — scores are the average of both: Model A more clearly delivers the product-ad look, with convincing steam, w | Anthropic: Claude Fable 5.1: Model A delivers the specified golden-hour look…
Mid-jump headphone sprint
A high-energy editorial sports image of a courier in a rust-red windbreaker leaping over a rain puddle while clutching a clear shopping bag containing boxed wireless earbuds, one sneaker kicked back and droplets exploding outward, motion blur in the trailing coat hem and spinning bicycle wheel behind them, dramatic low-angle composition under overcast city light, photorealistic style, 16:9.
OpenAI: GPT-6 Luna: Model B delivers a more energetic low-angle leap, dramatic spray, and a visibly spinning bicycle, while Model A has a convincing courier and clear shopping bag but feels more static. Neither clearly shows the courier wearing headphones, and Model B’s leap is less faithful to the specified kicked-back leg pose. (Second judge pass, order swapped — scores are the average of both: Model B more vividly captures the mid-jump action, dramatic low angle, rust-red jacket, clear earbud shopping bag, and e | Anthropic: Claude Fable 5.1: Model B delivers the dramatic low-angle, high-e…
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 better follows the restricted four-color, flat-vector treatment, with a clean and balanced campsite scene. Model A is richly composed but uses noticeable texture, shading, and a broader range of color variations that weaken adherence to the palette constraint. (Second judge pass, order swapped — scores are the average of both: Model B more clearly follows the restricted four-color flat-vector brief while depicting a recognizable dusk campsite with a balanced, legible layout. Model A is p | Anthropic: Claude Fable 5.1: Model B holds strictly to four flat colors (cre…
Rainy café checkout realism
A believable photorealistic scene at a neighborhood café checkout counter: a barista’s hand in natural skin detail passing a brushed stainless card reader across a worn walnut counter beside a folded navy canvas tote, a ceramic espresso cup with a thin crema ring, and a rain-speckled storefront window behind; soft cloudy daylight mixed with warm interior practicals, physically accurate reflections, shadow falloff, and convincing textures of metal, wood, skin, glazed ceramic, and fabric, 16:9.
OpenAI: GPT-6 Luna: Model B more clearly depicts the barista passing the reader in a rainy café, with the tote and espresso present in the scene. Model A has a more intimate, polished close-up, but the hand reads more as supporting the reader than handing it across the counter. (Second judge pass, order swapped — scores are the average of both: Model B more clearly captures the barista passing the reader across the counter and includes the espresso crema, navy tote, and rainy café window. Model A has a cleaner, hig | Anthropic: Claude Fable 5.1: Model A nails the specified materials—brushed s…
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 distinct espresso-filled cups in a clean straight row, closely matching the prompt. Model B also has seven clearly countable cups, but they are empty and blend more into the white marble. (Second judge pass, order swapped — scores are the average of both: Both images show exactly seven clearly separate cups in a straight overhead row on white marble. Model A makes the espresso cups more immediately recognizable with espresso inside and has a polished, appealing look, | Anthropic: Claude Fable 5.1: Both Model A and Model B render exactly seven cl…
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 gives a stronger close-up composition and more convincing mirror-like window reflections in the chrome, with a visible red apple reflection in the water. Model A is polished and attractive, but the required reflections in the water are less apparent. (Second judge pass, order swapped — scores are the average of both: Model B better matches the close-up teapot scene and shows the window and apple reflected across the chrome, with a red reflection visible in the water glass. Model A is cle | Anthropic: Claude Fable 5.1: Model A shows the window, the glass, and a clea…
Count eleven lip balms
A crisp top-down product photograph of EXACTLY eleven distinct lip balm tubes arranged on a pale terrazzo countertop, every tube fully visible and individually countable, with mismatched cap colors, tiny ingredient icons, and subtle cast shadows; clean commercial studio style with even diffused lighting and precise spacing, 16:9.
OpenAI: GPT-6 Luna: Model A presents exactly eleven fully visible, individually countable lip balms in a balanced two-row layout, with clear ingredient icons and subtle shadows. Model B also has eleven and renders readable labels, but its single row and large areas of empty space make the composition less balanced. (Second judge pass, order swapped — scores are the average of both: Model B presents exactly eleven fully visible tubes in a crisp, evenly spaced row, with distinct cap colors, ingredient icons, and espe | Anthropic: Claude Fable 5.1: Both Model A and Model B render exactly eleven…
Matchup powered by OpenRouter.