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
A prompt-by-prompt contest spanning precise object placement, reflective materials, named art styles, and elaborate scene-building. The results test not just visual polish, but whether each model can make a busy brief legible and stay faithful to its details.
Google’s Nano Banana Pro takes this matchup by a substantial margin: it wins seven of eight tasks, and its aggregate score leads FLUX 2 by 10.1 points. The limited confidence verdict backs up what the task results show—a consistent edge, not a single standout image carrying the comparison. That edge is clearest when prompts demand several constraints at once. Nano Banana Pro made the Moonforge relics countable and stuck to the requested inventory, while FLUX 2 added objects. It also gave the alchemist cabin a clearer, more balanced lineup and made the salvage-dock scene feel more cohesive, with the dismantled shuttle and wet reflections doing real compositional work. The win spans different kinds of image-making, too. Nano Banana Pro was judged stronger on the ukiyo-e wave and the geometric skywhale poster, and it produced the more convincing chrome teapot reflections. In the bedroom layout, it edged FLUX 2 overall, though placement details such as the lamp and rug drew criticism. These aren’t flawless generations; they’re examples of a model more often getting the whole brief into the frame. FLUX 2’s win matters: on attribute binding, it placed all four objects cleanly and made the requested matte-versus-glossy contrast especially clear. That result shows FLUX can be the better pick for a tightly specified, controlled product setup. But one task win cannot offset Nano Banana Pro’s broader strength across counting, scene construction, style, and material rendering. **Verdict: Google’s Nano Banana Pro wins decisively, 7–1.**
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 A cleanly presents all four objects in the requested positions, with clear matte-versus-glossy material contrast and soft studio lighting. Model B also follows the arrangement, but its tighter framing and visible banding and odd reflection on the sphere make it slightly less polished. (Second judge pass, order swapped — scores are the average of both: Model B more closely delivers the photorealistic tabletop look while clearly placing the duck atop the green cube and the blue cylinder behi | Anthropic: Claude Fable 5.1: Both place every object and color correctly (gr…
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 places the bed, centered round rug, desk beneath the back-wall window, and lamp in the front-right corner more precisely. Model B is polished, but the bed overlaps the rug and the lamp sits closer to the right wall than the front-right corner. (Second judge pass, order swapped — scores are the average of both: Model B more clearly places the bed against the left wall, centers the round rug, and puts the floor lamp in the front-right corner while maintaining a clean isometric style. Model | Anthropic: Claude Fable 5.1: Both place the bed on the left wall, a round ru…
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 shows the apple clearly and gives the teapot convincing reflections of the window and apple, with a stronger overall composition. Neither image clearly reflects both the window and apple in the water’s surface, but Model A also crops the apple heavily and makes the required reflections less legible. (Second judge pass, order swapped — scores are the average of both: Model B has a more convincing chrome reflection of the window and apple, with stronger close-up composition and dramatic li | Anthropic: Claude Fable 5.1: Model B delivers a far more photorealistic teap…
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 has more dynamic, boldly outlined waves and stylized foam, with a strong dawn atmosphere that closely evokes ukiyo-e; the extra boats are a minor departure from the singular subject. Model A follows the requested muted palette and single-boat setup well, but its rendering feels less graphic and its composition less forceful. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows fishing boats riding through a dramatic cresting wave, with bold outli | Anthropic: Claude Fable 5.1: Model B delivers traditional Japanese fishing b…
Bound Colors in the Alchemist Cabin
Inside a narrow fantasy alchemist cabin, a tidy workbench displays a hammered copper kettle, a matte cobalt-blue ceramic bowl, a translucent emerald glass vial, a pale ivory silk pouch, and a charred black iron key, each object clearly separated so its own color and material remain unambiguous, warm window light from the left, three-quarter view, crisp editorial product-style realism, 16:9
OpenAI: GPT-6 Luna: Model B presents all five specified objects clearly separated in a polished product-style arrangement, with warm light from the left and a richly realized alchemist cabin. Model A also follows the prompt closely, but Model B’s more balanced lineup and stronger sense of setting make the composition more complete. (Second judge pass, order swapped — scores are the average of both: Model B clearly separates all five requested objects and presents their colors and materials in a bright, warm-lit cab | Anthropic: Claude Fable 5.1: Model A nails the narrow cabin feel, left-side…
Neon Night Salvage Dock
A sci-fi salvage dock at neon night on the rain-slicked moon port of Kestra-9, where a lone mechanic in a patched pressure suit stands beneath flickering magenta and cyan signs beside a half-dismantled courier shuttle, puddles mirroring the glow, distant freighters drifting through haze, moody noir atmosphere, high-contrast wet reflections, dynamic wide composition, detailed cinematic concept art, 16:9
OpenAI: GPT-6 Luna: Model B more vividly conveys a working salvage dock, with a visibly dismantled shuttle, strong cinematic depth, and especially immersive wet reflections. Model A also matches the neon-noir setting well and clearly includes the moon and distant freighters, but Model B feels more detailed and compositionally cohesive. (Second judge pass, order swapped — scores are the average of both: Model B more fully captures the salvage-dock brief, with a visibly dismantled shuttle, patched-suit mechanic, read | Anthropic: Claude Fable 5.1: Model B delivers a clearly half-dismantled shut…
Moonforge Relics Count
A meticulously rendered fantasy still life on a black basalt altar inside the moonforge vault of Saint Veyra, showing exactly 11 distinct relics, all fully visible and individually countable: 3 cracked opaline masks, 2 silver astrolabes, 1 jade serpent idol, 1 brass censer, 1 folded indigo banner, 1 crystal dagger, 1 bone flute, and 1 sealed wax tablet; arranged in a single layer with no overlaps, centered composition, sharp studio clarity, soft edge shadows, cinematic realism, 16:9
OpenAI: GPT-6 Luna: Model B presents the requested relics in a clean, individually countable arrangement, while Model A appears to add an extra stone, astrolabes, and flute-like object. Model A has strong cinematic detail, but Model B better follows the exact inventory and single-layer composition. (Second judge pass, order swapped — scores are the average of both: Model B presents the specified 11 relics clearly in a tidy, countable single layer. Model A is visually rich, but includes extra flutes and an unrequest | Anthropic: Claude Fable 5.1: Model B renders exactly the 11 requested relics…
Melancholy Skywhale Poster
A clean stylized illustration of a melancholy skywhale drifting above the violet cloud canyons of Orun, tiny lantern gondolas hanging beneath its fins, with a lonely moon in the background; flat-shaded vector art with elegant geometric shapes, restrained linework, limited dusk palette of plum, teal, and cream, calm wistful mood, poster composition, 16:9
OpenAI: GPT-6 Luna: Model A captures the canyon setting and includes a clear gondola, while Model B more closely matches the clean geometric vector-poster style and gives the whale a distinctly melancholy expression. Model B’s hanging lights are less clearly gondolas and its canyon is more abstract, but its composition and polished graphic treatment make it the stronger overall fit. (Second judge pass, order swapped — scores are the average of both: Model B better captures the flat-shaded geometric poster style, re | Anthropic: Claude Fable 5.1: Model A delivers a cohesive poster composition…
Matchup powered by OpenRouter.