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 head-to-head across eight image-generation tasks tests prompt constraints, exact object counts, named-art-style control, and scene composition. The comparison also includes close calls where both models handled the brief well.
Google: Nano Banana Pro (Gemini 3 Pro Image) is the saved winner, taking four task wins to FLUX 2’s one, with three ties. Its aggregate score is **67.5**, against FLUX 2’s **64.9**. The saved editorial strength is **Clear result (not statistical)**; this assessment applies only to the saved task sample. Google’s edge came from a stronger salt-flats research scene, a more complete ukiyo-e composition, cups that were easier to identify and count, and a more convincing wary lynx. Those wins point to better execution on this set’s scene depth, visual specificity, and legibility—not a universal advantage. FLUX 2’s win was the negation task: it made the cozy reading nook feel more intimate while avoiding the prohibited decor. The moth macro, attribute-binding scene, and boathouse scene were ties, with each model offering strengths and neither establishing a decisive advantage across the judges’ assessments. **Final call: Google: Nano Banana Pro (Gemini 3 Pro Image), on this saved sample.**
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: Model A more directly captures the cozy nook with warm afternoon light, a steaming mug, and books arranged on the armchair, while avoiding the prohibited decor. Model B also follows the constraints, but its brighter, more cluttered arrangement feels less intimate. (Second judge pass, order swapped — scores are the average of both: Model B clearly includes the armchair, a stack of books, steaming tea, and a window while avoiding all prohibited items. Model A is also cozy and well composed, but pl | Anthropic: Claude Fable 5.1: Both Model A and Model B correctly omit plants,…
Salt Flats Research Convoy
A photorealistic wide-angle scene of a wildlife survey convoy crossing the cracked white Miraflores Salt Basin: in the foreground a compact yellow field truck with roof antennas, midground two biologists beside a tripod scope, farther back a line of six guanacos receding toward the horizon, and in the distance a rust-red observation tower and low mountains, all rendered with correct relative scale and a single consistent vanishing-point perspective along the tire tracks so every object feels properly sized and unwarped; crisp midday light, expansive sky, 16:9.
OpenAI: GPT-6 Luna: Model B more convincingly renders a photorealistic survey scene, with well-scaled foreground truck and researchers and clear depth across the salt flat. Model A has stronger crisp midday light, but its figures and equipment feel less natural and the guanacos recede less distinctly. (Second judge pass, order swapped — scores are the average of both: Model B more convincingly delivers the requested photorealistic salt-flat survey scene, with clear foreground-to-horizon depth, the biologists and sc | Anthropic: Claude Fable 5.1: Model B nails the layered depth the prompt asks…
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 evokes dawn with the sun and distant mountain, while its prominent fishing boat and dramatic wave create a stronger, more complete ukiyo-e composition. Model A is also highly faithful and has a restrained palette, but its boat is less visually distinct and the dawn setting is less explicit. (Second judge pass, order swapped — scores are the average of both: Model A presents a cleaner, more coherent ukiyo-e composition, with a clear dawn atmosphere and restrained indigo-and-c | Anthropic: Claude Fable 5.1: Both deliver convincing Great-Wave-style ukiyo-…
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: Both images show exactly seven cups in a straight row on white marble, but Model B makes the objects more clearly identifiable as cups with visible handles while keeping them distinct and countable. Model A is clean and orderly, though its handleless top-down view makes the cups slightly less unmistakable. (Second judge pass, order swapped — scores are the average of both: Model B more closely matches the requested even, soft-lit flat lay, with exactly seven clearly separate cups in a straight r | Anthropic: Claude Fable 5.1: Both Model A and Model B deliver exactly seven…
Dew Moth Macro
An ultra-detailed wildlife photograph of a glasswing moth perched on a bent stem of purple needlegrass at dawn, captured as an extreme macro close-up with a 100mm macro lens, razor-sharp focus on the moth’s translucent wing veins and the tiny dew beads clinging to its antennae, with the background dissolving into creamy green-and-amber bokeh; soft low sunrise light from the left, cinematic natural color, highly photorealistic, 16:9.
OpenAI: GPT-6 Luna: Model B presents a more immersive extreme macro with especially crisp dew, wing detail, and a beautifully rendered dawn bokeh; Model A has a graceful silhouette and strong translucent veins but feels slightly less detailed. Both capture the purple grass, soft left-side sunrise, and dew-beaded antennae well. (Second judge pass, order swapped — scores are the average of both: Model B has crisp macro detail and lovely dawn bokeh, but its wing looks more patterned and opaque than a glasswing’s. Mode | Anthropic: Claude Fable 5.1: Model B delivers a convincingly moth-like subje…
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 and Model B both place the correct colored objects in the requested arrangement, including the duck on the cube and the cylinder behind the foreground objects. Model B looks more convincingly photorealistic and uses the tabletop space effectively, while Model A has a cleaner studio backdrop and slightly more separation between objects. (Second judge pass, order swapped — scores are the average of both: Model A presents the requested left-to-right arrangement clearly, with the duck secure | Anthropic: Claude Fable 5.1: Both bind colors correctly and place the duck o…
Rain-Soaked Lynx Vigil
A realistic painted portrait of an Iberian lynx crouched beneath dark juniper branches during a cold rainstorm, its face turned slightly toward the viewer with unmistakable tense suspicion—narrowed amber eyes, flattened ears, whiskers pushed forward, mouth just barely parted—as droplets cling to its tufted fur and the pose communicates alert distrust rather than aggression; moody naturalistic style, subdued forest palette, shallow depth of field, side-lit by a break in storm light.
OpenAI: GPT-6 Luna: Model A clearly includes the overhead juniper and slightly parted mouth, but its more upright ears and open expression edge toward aggression. Model B better conveys a crouched, rain-soaked, wary lynx in a cohesive naturalistic painting, though the juniper framing and parted mouth are less explicit. (Second judge pass, order swapped — scores are the average of both: Model B conveys the tense, wary suspicion especially well, with convincing rain-soaked fur, a subdued storm palette, and highly nat | Anthropic: Claude Fable 5.1: Model B renders a convincingly Iberian lynx (sp…
Boathouse After the Catch
A photorealistic interior scene inside a weathered lakeside boathouse just after sunset: a park ranger in a moss-green wool sweater stands at a rough oak workbench cleaning a silver Arctic char beside a dented copper kettle, a coiled nylon net, wet leather gloves, and a brushed steel thermos, while a lantern hanging overhead casts warm light that mixes with cold blue twilight entering through slatted boards, producing believable soft shadows, specular highlights on metal, damp sheen on fish scales, realistic skin texture on the ranger’s hands, and convincing grain in the aged wood; composed like a candid documentary still, 16:9.
OpenAI: GPT-6 Luna: Model A more completely captures the candid boathouse scene, with the ranger, fish, requested props, and warm lantern light balanced against blue twilight. Model B has convincing textures and lighting, but its tight crop and awkwardly oriented, partly cut-off fish weaken the composition. (Second judge pass, order swapped — scores are the average of both: Model B more convincingly captures the candid, weathered boathouse mood, with especially effective warm lantern light against cool twilight and | Anthropic: Claude Fable 5.1: Model B delivers a genuinely photorealistic can…
Matchup powered by OpenRouter.