Head to head: FLUX 2 vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
FLUX 2 vs Google: Nano Banana 2 (Gemini 3.1 Flash Image)
By Ryan Merket · Published
Eight image-generation prompts put instruction following, composition, and visual finish under pressure—from counting identical cups to rendering reflections, named art styles, and a rain-soaked atelier.
Google: Nano Banana 2 (Gemini 3.1 Flash Image) takes all eight task verdicts, with no ties, and leads the aggregate score **68.7 to 56.1**. That makes this a decisive result on the saved task sample—not a claim about statistical confidence or performance beyond it. The advantage comes from repeatedly making complex scenes easier to read. Nano Banana 2 gives the alley fitting legible signage, steam, and puddle reflections; makes the window and apple reflections clearer in the teapot and water; and builds a more convincing rain-soaked atelier. It also handles the tabletop object relationships and the fishing boats in the ukiyo-e scene more directly. The wins are not flawless. Its seven cups are clearly countable but empty, and its Art Deco salon favors a broad-brimmed hat over the requested cloche. FLUX 2 has its own notable miss: the tight crop undercuts the requested full-body athlete portrait. It brings a clean studio look to that scene and a strong, restrained composition to the named-style task, but those strengths did not earn it a task win here. **Final call: Nano Banana 2 wins decisively on this saved task sample, combining clearer prompt execution with more convincing scene detail. FLUX 2 has moments of polish, but it falls behind across the full set of eight tasks.**
Hands on the laces
A full-body fashion portrait of a marathon tailor-athlete seated on a low locker-room bench, clearly tying the laces of a coral racing shoe with both hands visible in the foreground; accurate human anatomy is essential—natural shoulders, knees, feet, and especially correct hands with five believable fingers each, proper joint bends, realistic tension in the knuckles, and an unforced posture; wardrobe includes a slate merino base layer, running shorts, and a measuring tape draped around the neck, lit with clean softbox studio lighting against a simple backdrop.
OpenAI: GPT-6 Luna: Model B presents a more natural seated tying pose with both hands clearly working the coral shoe, and it strongly conveys the locker-room athlete setting with the measuring tape and running gear. Model A matches the clean studio styling, but its tight crop undermines the requested full-body portrait and makes the pose feel less natural. (Second judge pass, order swapped — scores are the average of both: Model B more convincingly shows the requested seated athlete tying a coral shoe, with the mea | Anthropic: Claude Fable 5.1: Model A crops off the head (not a full-body por…
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 espresso-filled cups, while Model B presents a larger, more evenly spaced and clearly countable row of seven identical cups. Model A has softer lighting, but Model B’s stronger composition and clarity give it the edge despite its empty cups and slightly harder shadows. (Second judge pass, order swapped — scores are the average of both: Model B presents exactly seven clearly separate, identical cups in a straight row on white marble, with a clean overhead composition. | Anthropic: Claude Fable 5.1: Both Model A and Model B deliver exactly seven c…
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 B looks more convincingly photorealistic as a tabletop scene, while keeping the duck atop the green cube and the blue cylinder behind the main objects. Model A has a cleaner studio backdrop, but feels more like a polished render. (Second judge pass, order swapped — scores are the average of both: Model B more convincingly delivers the photorealistic tabletop scene, with all requested objects and their color and placement relationships clearly visible. Model A follows the arrangement well, | Anthropic: Claude Fable 5.1: Both bind colors correctly and place the duck on…
Art Deco hat salon
An Art Deco poster-style illustration for the imaginary couture milliner 'Maison Vireline': an elegant model in a sharply tailored emerald skirt suit and sculptural cloche hat stands in a geometric hat salon with stepped mirrors, sunburst motifs, lacquered black displays, and gold inlay; the image must faithfully capture classic Art Deco aesthetics—streamlined forms, symmetrical composition, bold decorative geometry, limited jewel-tone palette, refined glamour, and stylized 1930s fashion illustration treatment—presented in a wide 16:9 layout.
OpenAI: GPT-6 Luna: Model B creates a more immersive, richly detailed Art Deco salon with stronger symmetry and a polished poster composition, though its hat is broad-brimmed rather than the requested cloche. Model A captures the cloche and renders the name cleanly, but its salon feels less cohesive and refined overall. (Second judge pass, order swapped — scores are the average of both: Model B more fully realizes the Art Deco salon with rich stepped mirrors, sunburst details, lacquered displays, gold inlay, and a | Anthropic: Claude Fable 5.1: Model B delivers a richer, more faithful Art Dec…
Neon alley fitting
A neon-night street-fashion fitting in a narrow back alley behind the fictional Saffron Circuit market: two models adjust an oversized reflective windbreaker on a dress form under a flickering magenta pharmacy sign and a cyan noodle-shop lantern, with puddles mirroring the colors and steam drifting from a vent; the mood should be unmistakably humid, electric, and midnight-urban, with deep contrast, colored rim light, and selective shadow shaping the scene, composed cinematically in 16:9.
OpenAI: GPT-6 Luna: Model B more fully captures the prompt with a clearly readable noodle-shop lantern, vivid pharmacy signage, reflective jacket fitting, steam, and puddle reflections in a richly layered alley. Model A has strong neon contrast and a clear fitting setup, but its signage is less legible and the alley feels less detailed. (Second judge pass, order swapped — scores are the average of both: Model B more fully realizes the humid, midnight-urban fitting with a detailed alley, vivid puddle reflections, st | Anthropic: Claude Fable 5.1: Model B delivers a richer, more immersive humid…
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 more clearly shows the window and red apple reflected in the chrome teapot, with window reflections also visible across the water’s surface; its brighter side light makes these details easier to read. Model A has a dramatic look, but the required reflections are less legible, especially in the water. (Second judge pass, order swapped — scores are the average of both: Model B more clearly shows the window and red apple reflected in the chrome teapot, with the window also visible through t | Anthropic: Claude Fable 5.1: Model B correctly shows the window, glass, and…
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 depicts fishing boats riding the waves at dawn, with a visible sunrise and convincing ukiyo-e linework and foam. Model A has a strong, clean composition and faithful muted palette, but its boat reads more like a larger vessel and the dawn setting is less explicit. (Second judge pass, order swapped — scores are the average of both: Model B more directly matches the fishing-boat subject and muted indigo-and-cream palette, with a dramatic, highly detailed ukiyo-e wave and dawn | Anthropic: Claude Fable 5.1: Model A delivers a clean, flat-planed Hokusai-st…
Rainy atelier realism
Photorealistic editorial scene inside a small bespoke outerwear atelier during a rainstorm: a copper clothing rack holding three waxed-cotton field jackets beside a walnut cutting table scattered with silver shears, tailor’s chalk, and a half-finished herringbone scarf; a patternmaker with damp skin and loose linen sleeves leans under a skylight, while raindrop reflections ripple across the polished concrete floor; physically convincing materials are critical—brushed copper, oiled walnut, wet glass, skin, wool, and crisp cotton all rendered with believable bounce light, soft shadow falloff, and subtle specular highlights, composed as a 16:9 documentary fashion photograph.
OpenAI: GPT-6 Luna: Model B better captures the rainstorm through the wet skylight, window, and reflective concrete, while presenting a convincing, richly detailed atelier scene. Model A includes the scarf and damp-looking craftsperson, but its floor reflections and workshop context are less developed. (Second judge pass, order swapped — scores are the average of both: Model B more fully captures the requested rain-soaked atelier, with convincing floor reflections, a visible rainy skylight, three jackets on a coppe | Anthropic: Claude Fable 5.1: Model B delivers a convincing documentary photo…
Matchup powered by OpenRouter.