Head to head: FLUX 2 vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
FLUX 2 vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
By Ryan Merket · Published
This image-model matchup spans tightly specified still lifes and posters as well as expressive, atmospheric scenes. The key divide is between exact adherence to a narrow brief and richer scene-building when the prompt asks for mood, style, or believable space.
Google: Nano Banana 2 (Gemini 3.1 Flash Image) leads the saved task sample with an aggregate score of **65.9** to FLUX 2’s **60.7**. It won six tasks, while FLUX 2 won two; there were no ties. That is a **clear result** in this sample—not a claim of statistical significance or a guarantee beyond it. Nano Banana 2’s advantage came from scenes where several visual demands had to work together. It handled the worried baker’s expression, sunken clafoutis, and exact clock time more completely, and made the attic workshop’s clutter and object scale more believable. Its woodblock-style dawn scene and rain-streaked conservatory also landed with stronger atmosphere and material detail. FLUX 2 was better when the brief rewarded disciplined execution over elaboration. It more faithfully arranged the hallway’s specified objects, and its poster kept the requested wording crisp in exactly three centered lines. Those wins point to a real strength in checklist-driven composition and layout, even as Nano Banana 2 more often combined prompt fidelity with a convincing overall scene. **Verdict: Google: Nano Banana 2 wins this head-to-head, 65.9 to 60.7.** The assessment applies only to the saved task sample: it favors Nano Banana 2’s richer, more coherent scene-making, while FLUX 2 remains the sharper choice in these particular tightly constrained 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: Model B looks more convincingly photorealistic, with a natural tabletop setting and all objects in the requested positions and colors. Model A has a cleaner studio arrangement, but appears more like a polished render than a photograph. (Second judge pass, order swapped — scores are the average of both: Model B better matches the photorealistic tabletop brief while clearly preserving the object colors and requested arrangement. Model A has a clean, readable composition, but its polished CGI appea | Anthropic: Claude Fable 5.1: Both correctly bind every color to its object 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 B more clearly conveys dawn with the rising sun and coastal horizon, while its dramatic wave, fishing boat, and woodblock-print details strongly match the prompt. Model A captures the indigo-and-cream palette and stylized foam, but its dawn setting is less evident and the boat is less compellingly integrated into the scene. (Second judge pass, order swapped — scores are the average of both: Model B more clearly centers a boat and its crew riding a dramatic cresting wave, with dawn scenery | Anthropic: Claude Fable 5.1: Model B delivers a convincing ukiyo-e print with…
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 has a dramatic campsite scene, but its blended sky and numerous tonal variations conflict with the strict four-color, no-gradient constraint. Model B uses a cleaner flat-vector style and a more consistent restricted palette while clearly depicting a campsite at dusk. (Second judge pass, order swapped — scores are the average of both: Model B follows the restricted four-color, flat-vector brief more closely and clearly conveys a campsite at dusk with a moon and stars. Model A has a strong | Anthropic: Claude Fable 5.1: Model B holds the four-color flat-vector constr…
Car-Free Umbrella Stand Hallway
An ultra-detailed editorial interior photograph of a narrow apartment entry hallway featuring an orderly still life of household objects only: a brass umbrella stand with four umbrellas, a low oak shoe bench with two pairs of boots, a framed print of orange sardines, a mail slot labeled "Flat 4C," and a striped runner rug leading to a frosted glass door at the end; the hall must be completely empty of people, pets, bicycles, and especially no cars or car imagery anywhere, soft morning light, clean straight-on composition, 16:9
OpenAI: GPT-6 Luna: Model A more closely matches the specified still life, including the four umbrellas, sardine print, labeled mail slot, and straight-on view of the frosted door. Model B has a strong, inviting hallway composition, but appears to show fewer than four umbrellas and adds several extra details. (Second judge pass, order swapped — scores are the average of both: Model A more clearly delivers the straight-on narrow hallway and the requested arrangement, including what appears to be four umbrellas and t | Anthropic: Claude Fable 5.1: Model A hits the checklist tightly (four umbrel…
Teapot Conservatory Reflections
A photorealistic still life on a walnut kitchen table inside a narrow rooftop conservatory at blue hour: a hammered copper teapot beside a half-filled ribbed glass carafe, two green figs, a folded saffron dish towel, and a tiny label reading "Morrow & Vale Preserves"; rain-speckled windowpanes behind them reveal a distant neon pharmacy sign and three dim apartment windows, all of which must appear plausibly reflected and refracted in the copper and the glass with correct distortion and transparency, including subtle reflections on the wet tabletop; cinematic side lighting, shallow depth of field, meticulous material rendering, 16:9
OpenAI: GPT-6 Luna: Model B has stronger cinematic composition and more convincing rain, material detail, and distorted reflections through the glass and across the copper. Model A includes the requested separate label and all key props, but appears darker and less polished; Model B’s label placement and text are less exact. (Second judge pass, order swapped — scores are the average of both: Model B better captures the cinematic, rain-lit conservatory with especially convincing distorted neon reflections in the cop | Anthropic: Claude Fable 5.1: Model B delivers a more convincing photoreal sc…
Worried Baker Portrait
A pastel storybook illustration of a middle-aged home baker in a small midnight kitchen, clutching an oven mitt to her chest while staring at a sunken blackberry clafoutis on the counter with unmistakable worried embarrassment: knitted brows, tight mouth, slightly raised shoulders, and watery eyes; around her are domestic details like a chipped cream mixing bowl, a tin marked "Dunley Flour No. 3," and a crooked wall clock at 12:17, all arranged in a cozy triangular composition under one warm pendant lamp and cool moonlight from a side window, 16:9
OpenAI: GPT-6 Luna: Model B more fully captures the worried embarrassment, visibly sunken clafoutis, exact 12:17 clock, and requested kitchen details, with a strong cozy triangular composition. Model A is appealing but its clock is not 12:17 and the clafoutis does not look sunken. (Second judge pass, order swapped — scores are the average of both: Model B more fully captures the prompt with a strongly worried expression, a visibly sunken berry clafoutis, moonlit window, warm pendant, and the specified kitchen detai | Anthropic: Claude Fable 5.1: Model B nails the emotional brief (watery eyes,…
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 A follows the requested three-line centered layout with crisp, correctly spelled text on a deep navy, subtly grained background. Model B renders the wording legibly but splits the headline across two lines and adds a photographic setting that departs from the minimalist poster brief. (Second judge pass, order swapped — scores are the average of both: Model A presents the exact text in three centered lines on a deep navy, subtly grained background, while Model B breaks the headline across t | Anthropic: Claude Fable 5.1: Model A delivers exactly three crisp, correctly…
Attic Workshop Scale Study
A realistic wide-angle view of a cluttered attic workshop in one-point perspective, with all lines consistently converging to a single vanishing point at a small dormer window: in the foreground a large blue enamel washbasin and a wooden crate of clothespins, in the midground a sewing machine on a worktable with jars of buttons, and in the background a tiny toy sailboat on the windowsill beside a much taller lamp, all objects sized believably relative to one another and the room; dusty sunbeams, crisp textures, balanced asymmetrical composition, 16:9
OpenAI: GPT-6 Luna: Model B offers a more convincingly scaled, richly cluttered workshop with clear one-point perspective and all requested objects. Model A has more dramatic dusty sunbeams, but its oversized basin dominates the scene and makes the room’s scale feel less believable. (Second judge pass, order swapped — scores are the average of both: Model B more convincingly delivers a cluttered, believable attic workshop, with the requested objects arranged at more plausible relative scales and the room’s perspect | Anthropic: Claude Fable 5.1: Model B nails the scale relationships central t…
Matchup powered by OpenRouter.