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
We compared FLUX 2 and Nano Banana Pro across eight image-generation tasks, from exact layouts and readable text to anatomy and named art styles. The results reveal where each model is dependable—and where prompt fidelity matters more than polish.
Google’s Nano Banana Pro takes the matchup, 64.9 to 60.0, winning five tasks to FLUX 2’s two, with one tie. The limited confidence rating makes this a clear overall result, not a sweep: FLUX has two standout wins, and some individual tasks were close. Nano Banana Pro’s advantage is in following visual instructions precisely. It won the lobby composition, showed the bracelet-making hands more convincingly, and delivered the stronger ukiyo-e treatment. It also rendered the wayfinding poster with the exact wording on the requested three lines—a meaningful edge when legibility and layout are the task. In the counting test, both models produced seven cups, so that result is a much less persuasive separator than the poster win. FLUX 2 was better at conveying a specific moment. Its conservator reads as acutely anxious rather than merely concerned, and its atrium jumper gets more kinetic force from the suspended ring and flying admission stubs. The bedroom-layout task ended in a tie: each image captured important spatial relationships, while each missed some details. **The call: Nano Banana Pro wins on more consistent prompt execution, especially for style, text, and composed scenes. FLUX 2 is the stronger pick for these particular high-drama moments—but across this set, Google’s model is the more reliable image-maker.**
Exact Lobby Layout
An ultra-detailed editorial photograph of a boutique hotel lobby arranged with exact spatial relationships: a tall matte-black floor lamp stands to the left of a rust-colored velvet armchair; a narrow travertine side table is directly between the lamp and the chair; a small bronze fox sculpture sits on top of the side table; behind the chair is a large potted bird-of-paradise plant; to the right of the chair is a cream oval rug with a closed teal suitcase resting on it; in front of the chair is a low glass coffee table holding exactly two stacked coral books; soft morning window light, balanced straight-on composition, 16:9.
OpenAI: GPT-6 Luna: Model B presents the requested objects in a clearer, more balanced straight-on layout, with soft window light and the two coral books legible on the glass table. Model A includes nearly all the same elements, but its coffee table is heavily cropped and its warmer lighting departs more from the prompt. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the specified spatial relationships and the requested 16:9, straight-on editorial composition, with | Anthropic: Claude Fable 5.1: Model A satisfies every spatial constraint, incl…
Anxious Conservator
A close, realistic oil painting of a restoration conservator in an ornate townhouse library, pausing on a rolling ladder with one gloved hand clutching a tiny cracked porcelain fragment, her face vividly conveying anxious dread—brows drawn upward, eyes glossy and fixed downward, lips parted in a tense inhale, jaw tight; warm lamplight mixes with cool dusk from a tall window, carved shelves and faded mural details framing her, shallow depth of field focused on the expression.
OpenAI: GPT-6 Luna: Model A more closely matches the requested tight, expression-focused scene, with vividly anxious features, a gloved hand holding the cracked fragment, and the ladder framed by warm lamplight and cool dusk. Model B is a polished oil-painting interpretation with strong restoration details, but its wider composition gives less emphasis to the face and the expression reads as worried rather than acute dread. (Second judge pass, order swapped — scores are the average of both: Model B better captures | Anthropic: Claude Fable 5.1: Model A delivers a convincingly realistic oil-pa…
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 A gives a tighter view with soft daylight and visible hands, but the fingers and thread grip are less clearly defined. Model B more convincingly shows both hands working the bracelet in a natural candid setting, with stronger overall photographic detail. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows a candid bracelet-making action, with both hands gripping the colorful threads in a plausible way. Model A has a pleasing close-up and soft ligh | Anthropic: Claude Fable 5.1: Model B delivers a convincing candid scene with…
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 convincingly captures ukiyo-e with bold outlines, flat indigo-and-cream planes, stylized foam, and a dramatic fishing boat beneath the cresting wave. Model A has a strong dawn palette and wave, but its gradients and detailed sailing vessel feel less faithful to the requested woodblock tradition. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows a fishing boat riding the cresting waves, with fishermen, a dawn horizon, and the requested ind | Anthropic: Claude Fable 5.1: Model B is far more faithful to the ukiyo-e tra…
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 clearly separated, identical espresso cups in a straight row on white marble, with espresso visibly inside and balanced overhead framing. Model B also has seven cups in a straight row, but they are empty and the row sits noticeably low in the composition. (Second judge pass, order swapped — scores are the average of both: Model B shows exactly seven clearly separate, identical cups in a clean straight row with soft, even lighting and visible handles. Model A also has | Anthropic: Claude Fable 5.1: Both Model A and Model B render exactly seven se…
Atrium Leap
A contemporary museum atrium during a kinetic dance rehearsal, a single dancer in a cobalt jumpsuit captured mid-leap from a low concrete stair toward a suspended brass ring, one foot tucked under, the other extended, hair and fabric trailing to show real momentum; visitors blur in the background and a scattering of paper admission stubs lifts in the air wake, emphasizing motion; architectural photography style with crisp realism, dramatic side lighting from a skylight, wide 16:9 composition from a slightly low angle.
OpenAI: GPT-6 Luna: Model A more fully captures the prompt with a suspended brass ring, a vivid mid-leap pose, abundant airborne admission stubs, and dramatic skylight lighting. Model B has a spacious, polished atrium and a convincing leap, but its ring appears freestanding and the paper and motion cues are less pronounced. (Second judge pass, order swapped — scores are the average of both: Model A better fulfills the suspended-ring and kinetic-motion details, with abundant paper stubs, trailing hair and fabric, an | Anthropic: Claude Fable 5.1: Model A nails the suspended brass ring, the dra…
Readable Wayfinding Poster
A clean Scandinavian interior corridor with a wall-mounted wayfinding poster designed in bold modern typography, photographed straight-on so every letter is easy to read; the poster must clearly say exactly: "NORTH ANNEX" on the first line, "ROOM 7B" on the second line, and "QUIET ARCHIVE" on the third line; charcoal text on a pale sand background, minimal graphic design, even gallery lighting, uncluttered composition, 16:9.
OpenAI: GPT-6 Luna: Model B renders the exact requested wording on three readable lines, while Model A wraps “NORTH ANNEX” onto two lines and therefore misses the specified layout. Both are clean, polished corridor scenes, though Model B’s poster is especially prominent and easy to read. (Second judge pass, order swapped — scores are the average of both: Model B presents all three exact phrases on the requested lines in a clear, straight-on corridor composition. Model A is polished, but splits “NORTH ANNEX” across | Anthropic: Claude Fable 5.1: Model B renders the text exactly as specified on…
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 more clearly places the bed along the left wall, centers the round rug, and puts the desk beneath the back-wall window with the lamp in the front-right corner. Model B is a clean isometric illustration too, but the bed overlaps much of the rug and the lamp sits farther back along the right wall. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the requested clean flat-vector style while clearly placing the bed, centered round rug, desk bene | Anthropic: Claude Fable 5.1: Model A nails the bed flush against the left wa…
Matchup powered by OpenRouter.