Nano Banana Pro and Nano Banana 2

What it is

Nano Banana Pro is Google’s in-house codename for Gemini 3 Pro Image, the upper half of a two-model image family. Its companion, Nano Banana 2, is Gemini 3.1 Flash Image and sits below it at an Elo of 1254. The names are light-hearted, the positioning less so: the Pro tier is tuned for quality and fine control, while the Flash tier is tuned for speed and volume. On the image-editing arenas, the leaderboards that measure how well a model alters an existing picture rather than inventing one from scratch, Nano Banana Pro takes the lead, and it holds that lead even on the arenas where OpenAI’s GPT Image 2 tops the generation charts.

Editing versus generation

That split between editing and generation is the useful thing to grasp here. Generating an image means producing one from a text prompt alone. Editing means taking a supplied photograph or render and changing part of it, swapping a background, recolouring an object, adjusting a pose, while keeping the rest intact and believable. Editing is the harder discipline for most working teams, because production work usually starts from an existing asset rather than a blank canvas, and it is exactly the ground where the editing champion earns its title.

The technical reason is the family’s handling of physics-accurate materials and lighting. Reflections, textures, shadows and the way light falls across a surface are where edited images most often betray themselves, and the Pro model keeps them consistent. That is why product photographers and e-commerce teams reach for it: a system that renders glass, metal, fabric and skin correctly can stand in for an expensive studio shoot across a large share of catalogue work, and do it in seconds rather than days.

Pricing and access

Pricing is a second draw. The family is the cheapest credible route to genuine 4K output, at about $0.24 for a full-resolution image, and a batch mode roughly halves the 2K rate to around $0.067 per image for teams generating in bulk. Speed is covered by a third variant, Nano Banana 2 Lite, which shipped on 2 July 2026 and produces images in under four seconds from about $0.034 per thousand, cheap enough to sit inside interactive tools where a user waits for the result.

Access is broad. The weights are closed, so the models cannot be downloaded and run privately, but they are reachable through the Gemini app for casual use, through Google AI Studio for prototyping, and through the Gemini API for production, with a free tier that lets a team judge the output before committing any spend. That mix of a no-cost entry point and low per-image pricing at the top end is unusual, and it lowers the barrier to trying Nano Banana Pro against an incumbent before a single invoice is raised.

The three tiers

The lineup is best read as a single strategy expressed across three points. The Pro model answers the demand for maximum quality and editing control, Nano Banana 2 covers the middle ground where its Elo of 1254 still buys competitive output at lower cost, and the Lite variant takes the floor on price and latency. A team can therefore stay within one family as a job moves from a carefully composed hero image to thousands of routine variations, without learning a new tool or migrating between suppliers at each step. Few rivals present so tidy a ladder from premium quality down to high-volume speed.

What to watch

Within the broader set of image models, the family occupies a clear position: the editing champion, and the value leader at high resolution. The fair qualification is that Google does not lead every arena at once, since the generation honours currently sit elsewhere, but for the everyday work of changing real images rather than dreaming up new ones, Nano Banana Pro is the reference point. What to watch is whether rivals can match its grasp of materials and lighting, and whether the Flash and Lite tiers pull more real-time and on-device work towards Google as prices keep falling.