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
Eight image-generation tasks put these models through strict style constraints, precise layouts, typography, and demanding hand-and-object details. The results show where each model is dependable—and where attractive images still miss the brief.
Google’s Nano Banana Pro wins this matchup on more than a narrow scoring quirk: it leads 66.3 to 54.0, takes six task wins, and loses none. The evaluation puts that advantage at limited confidence. FLUX 2 has appealing images, but repeatedly gives up prompt fidelity for polish or extra detail. The clearest gap appears in constraint-heavy work. On the restricted-palette campsite, FLUX 2’s gradients and added color variation broke the four-color rule; Google’s flatter shapes kept the palette disciplined and the scene legible. Google also came out ahead on spatial layout and the accessory arrangement, where clean separation and faithful placement mattered more than decorative finish. Hands were another meaningful separator. Google’s bracelet image better showed the candid act of tying, with loose threads and a convincing working grip; its tailor portrait also kept both hands visible holding the tape. FLUX 2’s hands were often serviceable, but its bracelet looked finished rather than in progress, and its tailor image cropped the face and wrapped the tape around the torso. FLUX 2 did have strengths: its campsite was rich and balanced, and its showroom mannequin better conveyed claymation, even though Google won that task overall. The models tied on the reading nook and boutique poster, where both handled the core brief but each had trade-offs. Those ties don’t erase the pattern across the other six tasks: **Google: Nano Banana Pro is the clear overall winner; FLUX 2 is capable, but less reliable when the prompt gets specific.**
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 rich, balanced campsite scene, but its gradients and extra color variations conflict with the strict four-color palette. Model B uses a cleaner, flatter palette and crisp vector shapes, with a bold composition that makes the campsite immediately legible. (Second judge pass, order swapped — scores are the average of both: Model B closely follows the four-color, flat-vector restriction and presents a clear, balanced campsite scene. Model A has an appealing, detailed composition, but | Anthropic: Claude Fable 5.1: Model A delivers an attractive scene but violate…
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 furnishings very accurately, especially the lamp in the front-right area, while Model B is more polished and better matches the requested flat-vector style. Model B’s lamp sits closer to the side wall than the front-right corner, but the overall layout remains clear and consistent. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the requested flat-vector style and presents the bed, centered round rug, desk beneath the back-wall | Anthropic: Claude Fable 5.1: Model A places the bed, rug, and desk-under-wind…
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: Both images include the requested nook elements and avoid plants, lamps, and wall art. Model B has a more complete, balanced wide composition, with the chair, books, mug, and sunlit window all clearly visible. (Second judge pass, order swapped — scores are the average of both: Model B clearly includes the requested chair, books, mug, and window while respecting all exclusions; Model A does too, with the tea and cozy throw especially clear. Model A’s closer, warmer composition feels more inviting | Anthropic: Claude Fable 5.1: Both Model A and Model B include the armchair,…
Precise showroom layout
A stylish stop-motion claymation fashion showroom scene viewed at eye level, soft directional afternoon light, 16:9; place a coral mannequin wearing a denim jacket exactly in the center, a mustard shoe box on top of a low white pedestal, the pedestal to the left of the mannequin, a mint scarf draped behind the mannequin, a tiny chrome handbag on the floor to the right of the mannequin, and a black hat positioned between the pedestal and the handbag; the spatial relationships must be unambiguous and exactly as described, with a tidy boutique interior and shallow depth of field.
OpenAI: GPT-6 Luna: Model A more clearly conveys the stop-motion claymation style and presents a complete coral mannequin centered among the requested accessories; Model B follows the layout well but uses a headless mannequin and feels less distinctly claymation. Both place the pedestal left, scarf behind, handbag right, and hat between the pedestal and handbag. (Second judge pass, order swapped — scores are the average of both: Model B follows the requested layout especially clearly, with the mannequin centered an | Anthropic: Claude Fable 5.1: Both place every object in the correct spatial…
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 B better captures the candid act of tying a bracelet, with loose threads and a convincing working grip in soft outdoor light. Model A has clearly visible, mostly natural hands, but the bracelet looks more finished and the image feels less candid. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows the hands actively tying a colorful friendship bracelet, with loose threads and a natural candid setting. Model A has a clean, balanced close-up, but th | Anthropic: Claude Fable 5.1: Model B delivers a far more convincing candid t…
Typography boutique poster
A high-end fashion boutique window poster in a clean Swiss graphic design style, photographed straight-on in crisp morning light, 16:9; the poster must be perfectly legible and easy to read, with these exact lines of text: "VELORA THREAD HOUSE" at the top, "Sample Sale — 19 OCT" beneath it, and "Silk coats from $84" at the bottom; use elegant black sans-serif type on warm ivory paper, with a small illustration of a teal trench coat and copper sewing shears, but keep the text dominant, sharp, and correctly spelled.
OpenAI: GPT-6 Luna: Model A presents the exact wording clearly in a balanced, nearly straight-on poster with the requested coat and shears, though its type is serif rather than sans-serif. Model B uses sans-serif type, but the angled view and lowercase “oct” make it less faithful to the prompt’s exact text and presentation. (Second judge pass, order swapped — scores are the average of both: Model B better follows the clean Swiss style with legible black sans-serif typography and a restrained illustration, though th | Anthropic: Claude Fable 5.1: Both render every line of text correctly and le…
Tailor with accurate hands
A realistic editorial portrait of a young bespoke tailor in a sunlit atelier, standing three-quarters toward camera beside a dress form, natural proportions, medium shot, 16:9; the tailor is wearing a charcoal vest over a pale blue shirt and holding a yellow measuring tape with both hands fully visible in front of the torso, each hand anatomically correct with five natural fingers, clean joints, and believable pose, while the face, arms, and posture remain proportionate and relaxed; scissors, chalk, and folded fabric sit softly out of focus on the worktable behind.
OpenAI: GPT-6 Luna: Model B more closely matches the requested atelier portrait, with both hands holding the tape in front of the torso and a clear dress form and worktable; the hands look more natural overall. Model A has a convincing tailor’s outfit and setting, but the tape is wrapped around the torso and the face is more severely cropped. (Second judge pass, order swapped — scores are the average of both: Model B better matches the three-quarter editorial portrait and sunlit atelier, with both hands visible hol | Anthropic: Claude Fable 5.1: Model A crops the face out entirely and has the…
Color-material accessory binding
A meticulously arranged fashion flat lay shot from directly above on a pale concrete surface, editorial studio lighting, 16:9, showing six accessories with strict color and material binding: a matte cobalt leather handbag, a glossy amber acrylic bangle, a brushed silver metal belt buckle, a forest-green velvet beret, a cream ceramic perfume bottle, and a translucent plum glass sunglasses case; each item must clearly keep its own specified color and material without any colors swapping or bleeding onto the wrong object, with realistic textures and neat spacing.
OpenAI: GPT-6 Luna: Model B presents all six accessories with clearer separation and more convincing material textures, while keeping the requested colors distinct. Model A is polished, but its silver buckle appears awkwardly entangled with the amber bangle, making the arrangement less clear. (Second judge pass, order swapped — scores are the average of both: Model B presents all six requested accessories with the specified colors and materials, clean spacing, and no apparent color or object confusion. Model A is a | Anthropic: Claude Fable 5.1: Model B renders exactly six cleanly separated i…
Matchup powered by OpenRouter.