Head to head: FLUX 2 vs Google: Nano Banana Pro (Gemini 3 Pro Image)
FLUX 2 vs Google: Nano Banana Pro (Gemini 3 Pro Image)
By Ryan Merket · Published
This image-model matchup puts tightly specified scenes, layouts, and palettes alongside a demanding text-rendering task. The comparison turns on how reliably each model follows those constraints while still producing a polished image.
Google: Nano Banana Pro (Gemini 3 Pro Image) is the saved winner, with an aggregate score of **65.5** to FLUX 2’s **58.4**. It won seven of the eight tasks; FLUX 2’s sole win was the multi-line text poster, where it kept the requested wording in exactly three centered lines while Google broke the title across lines. Google’s edge came from repeatedly getting the requested relationships and visual constraints right: it staged the pantry tableau more faithfully, built a more cohesive isometric room, and delivered stronger material detail in the rainy sink scene. It also better respected the four-color limit in the campsite and kept the specified colors and materials clear in the breakfast still life. Its Vienna kettle poster is the saved task winner, despite FLUX’s stronger match on some individual details. FLUX 2 was not without strengths: its stop-motion character work stood out in the pantry scene, and its exact three-line typography was the clearest task-specific win. But those successes did not offset Google’s broader consistency across composition, scene detail, and constraint-following in this set. **Final call: Google: Nano Banana Pro (Gemini 3 Pro Image), a decisive result on this task sample—not a statistical claim. This assessment applies only to the saved task sample.**
Pantry negotiation tableau
A coherent stop-motion claymation kitchen-counter scene where seven household objects interact like characters: a tall glass bottle of sunflower oil leaning toward a blue ceramic mixing bowl, a wooden rolling pin bridging between two chipped mugs, a lemon perched on a folded dish towel, a small tin sieve dangling from the bowl’s rim, and a box of saffron threads tipped open so a few strands spill forward; arrange them in a balanced triangular composition with clear depth, believable contact points, and warm late-afternoon window light across the whole scene, 16:9.
OpenAI: GPT-6 Luna: Model B closely matches the requested object placements and warm kitchen-counter tableau, while Model A has a stronger stop-motion claymation look and expressive character faces but misses several specified contacts and arrangements. Model B is more compositionally precise and technically polished, though it feels more photorealistic than claymation. (Second judge pass, order swapped — scores are the average of both: Model B more faithfully stages the requested relationships, including the oil l | Anthropic: Claude Fable 5.1: Model B satisfies nearly every spec—oil bottle…
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 A follows the requested layout cleanly, while Model B includes the same key placements with a more inviting, cohesive flat-vector style and a stronger sense of room detail. Both maintain a consistent isometric perspective, though Model A is slightly more restrained and uncluttered. (Second judge pass, order swapped — scores are the average of both: Both images place the bed, centered round rug, desk beneath the back-wall window, and floor lamp in the requested locations. Model B has a more | Anthropic: Claude Fable 5.1: Both place the bed left, rug centered, and desk…
Rainy sink realism
A photorealistic close view of a real kitchen sink at dusk: a brushed stainless-steel faucet with tiny water spots above a wet walnut cutting board, a crumpled off-white linen dishcloth, a sliced blood orange with glossy flesh, and a human hand reaching in from the right to set down a heavy dark-green glass bottle; cool rain-filtered daylight from a nearby window mixes with a weak under-cabinet bulb, creating physically convincing reflections, soft shadow falloff, subsurface scattering in the citrus, damp fabric texture, realistic skin, and accurate metal, wood, glass, and cloth materials, 16:9.
OpenAI: GPT-6 Luna: Model B more completely captures the rainy sink scene, with convincing faucet details, wet wood, textured cloth, glossy citrus, and the warm-cool mixed lighting. Model A is attractive but has a less convincing faucet and weaker material detail overall. (Second judge pass, order swapped — scores are the average of both: Model B more convincingly captures the rainy, dusk-lit kitchen, with the requested spotted faucet, wet walnut board, cloth, blood orange, and hand placing a green bottle all clear | Anthropic: Claude Fable 5.1: Model A delivers the requested close view with…
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 presents the exact three requested lines, centered and legible on a deep navy background. Model B renders the wording correctly but breaks the title into two lines, departing from the requested three-line layout and making the poster feel less balanced. (Second judge pass, order swapped — scores are the average of both: Model A presents the requested wording in exactly three crisp, centered lines, while Model B breaks the title across two lines and creates a four-line layout. Both use a | Anthropic: Claude Fable 5.1: Model A renders all three lines exactly as speci…
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 delivers a very clear, centered one-point aisle, but its soft detail and repetitive lighting feel less photorealistic. Model B keeps the shelves and floor converging naturally toward a single endpoint while offering richer, more convincing detail and warm illumination. (Second judge pass, order swapped — scores are the average of both: Model A more clearly emphasizes a centered one-point perspective, with the shelving and books consistently diminishing toward a single distant vanishing p | Anthropic: Claude Fable 5.1: Model A delivers a clean, centered one-point pe…
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 palette and flat-vector requirement, while Model A uses prominent gradients and many tonal variations. Model A has a richly layered composition, but Model B is clean, balanced, and clearly depicts a mountain campsite at dusk. (Second judge pass, order swapped — scores are the average of both: Model B closely follows the four-color, flat-vector constraint and presents a clear, balanced campsite-at-dusk scene. Model A has a strong, detailed composit | Anthropic: Claude Fable 5.1: Model A has a polished composition but violates…
Vienna Secession kettle poster
A Vienna Secession–style illustration of a polished copper tea kettle on a narrow marble shelf, flanked by three square spice tins and a single striped pear, rendered with flat ornamental geometry, disciplined gold linework, stylized botanical motifs, and a muted palette of olive, cream, rust, and black; the composition should feel like a 1903 exhibition poster with elegant negative space, symmetrical framing, and a decorative sunrise window behind the objects, 16:9.
OpenAI: GPT-6 Luna: Model A more closely matches the polished copper kettle, veined marble shelf, muted palette, sunrise, and intricate botanical framing. Model B has convincing exhibition-poster typography and a strong Secession feel, but its shelf is dark rather than marble and its kettle reads as less polished copper. (Second judge pass, order swapped — scores are the average of both: Model A has a more disciplined, symmetrical poster composition, a clearer sunrise window, and a stronger match to the muted palet | Anthropic: Claude Fable 5.1: Model B captures the Vienna Secession idiom far…
Color-locked breakfast still life
A precise still life on a pale stone tabletop featuring exactly six separate objects, each with its own bound color and material: a matte cobalt-blue ceramic teapot, a glossy mustard-yellow glass tumbler, a brushed copper spoon, a soft mint-green linen napkin, a dark plum-purple velvet ring box, and a chalk-white plaster egg cup; arrange them in a clean, slightly staggered composition under crisp morning side light so every object's color stays strictly on the correct object, with no swapping, bleeding, or material confusion, 16:9.
OpenAI: GPT-6 Luna: Model A presents all six objects with their specified colors and materials in a tighter, more balanced arrangement, while Model B leaves more empty space and has harsher shadows. Both render the objects clearly and follow the prompt closely. (Second judge pass, order swapped — scores are the average of both: Model B better preserves the specified matte ceramic teapot and clean separation of all six correctly colored objects, with crisp morning side light and a pleasing staggered layout. Model A | Anthropic: Claude Fable 5.1: Both renders deliver all six objects with correc…
Matchup powered by OpenRouter.