
Pro quality at flash speed. 4-8 seconds per image.
Built on Google's Gemini 3.1 Flash Image, Nano Banana 2 is an autoregressive model that reasons about composition, lighting, and spatial relationships before rendering. Unlike diffusion models, it plans each image through a three-stage loop: plan, evaluate, improve. The result is accurate text rendering, consistent multi-person scenes, and reliable spatial composition at roughly half the cost and twice the speed of the Pro variant.

“A sleek electric sports car in matte grey, studio lighting, three-quarter front view on a dark reflective surface”

“A dark fantasy RPG environment: crumbling stone archway overgrown with bioluminescent vines, volumetric fog”

“Flat lay product photography of a skincare range on travertine stone, soft overhead light, editorial composition”

“A modernist villa on a hillside at golden hour, warm timber cladding, floor-to-ceiling glazing reflecting the sky”

“An exhibition stand concept for a furniture brand: open timber frame structure, pendant lighting, neutral palette”

“A sleek laptop on a dark desk displaying a modern website UI with bold typography and orange accents, moody studio lighting, editorial tech photography”
Type a detailed prompt or upload a reference sketch, photo, or mood board.
Pick your resolution and aspect ratio. See the credit cost before you generate.
Your image is delivered in seconds. Download, iterate, or pipe into video.
Jump into the Studio and start generating. Plans from £10/month.
Choose a PlanNano Banana 2 runs on Google's Gemini 3.1 Flash Image architecture, a fundamentally different approach to image generation. Where traditional diffusion models build images by denoising random noise, Nano Banana 2 is autoregressive: it predicts visual tokens through the same reasoning pipeline that handles text. This means it genuinely understands your prompt rather than matching weighted keywords.
For design professionals across every discipline, this matters. Whether you are describing a vehicle concept with specific surfacing and proportions, a game environment with volumetric lighting, or a product flat lay with precise composition, the model reasons about what you mean before committing pixels. It verifies spatial relationships, checks quantities, and validates typography character by character. Fewer re-generations, output you can actually use.
Nano Banana 2 generates at native resolution across 10+ aspect ratios, from ultra-wide 21:9 panoramas to 9:16 portrait formats for social content. Output at 4K is native, not upscaled. Generation takes 4-8 seconds at 1K, making it fast enough for real-time concept iteration during a client meeting or design crit.
One of Nano Banana 2's standout capabilities is typography. It validates text character by character across multiple languages, producing legible signage, labels, and branded content. Whether you need text on a storefront, a product label, a game UI overlay, or an exhibition graphic, it will be readable.
At 4-8 seconds per generation, Nano Banana 2 is designed for rapid concept exploration. Test 10 variations of a vehicle colourway, explore material palettes for a product range, iterate on a character design, or draft social content at volume before committing to a premium render with Nano Banana Pro.
Supports up to 14 reference images per generation. Upload sketches, photos, material samples, mood boards, or existing renders alongside your prompt. The model uses these as visual context, not just style references. Feed it a hand-drawn concept and get back a production-quality visualisation that respects your original intent.
A small indie studio building creative tools the way they should be built. No VC theatre, no funnel games, no faceless support.
Professional image generation. Plans from £10/month.
Also available on Stensyl