Model · Google

Nano Banana, on a canvas built for production

Nano Banana is what everyone calls Google's image models. Its signature is one thing almost no other model does cleanly: it writes legible text inside an image. On Beemm Vision its five variants run in the same canvas as the rest of the catalogue, on a single credit balance.

What this model does better than the rest

Text that actually reads

This is the headline feature. Shop signs, posters, product labels, street signage: Nano Banana renders type without mangling it, accents included. The image above received no text layer — the gold lettering came out of the model.

Editing by conversation

You edit an existing image in plain language: change the light, remove an object, replace a background, restyle a character. No mask to trace, no layer stack. Inside a Beemm Vision workflow that edit becomes a node you replay across a hundred images.

The same character across shots

Give it a reference photo and the subject stays recognisable from scene to scene. That is what makes a storyboard, a campaign series or a product shot in ten different settings actually usable. Nano Banana 2.1 takes up to fourteen reference images when editing.

Up to 4K, from 8:1 to 1:8

The Pro, 2 and 2.1 variants output at 1K, 2K or 4K. For text-to-image, Pro offers ten aspect ratios, from a 21:9 panorama to a 9:16 vertical; Nano Banana 2 and 2.1 offer fourteen, up to the extreme 8:1 and 1:8 formats — a banner and its story format without cropping either.

The five variants available here

You switch between them in the model picker while generating — not through a separate subscription.

  • Nano Banana 2.1The newest. Text-to-image, or editing from one to fourteen reference images, at 1K, 2K or 4K, fourteen formats from 8:1 to 1:8.
  • Nano Banana 2Text-to-image and image-to-image, 0.5K up to 4K.
  • Nano Banana ProThe quality tier: same approach, a more finished render, 1K / 2K / 4K.
  • Nano Banana 2 LiteFast and cheap at 1K. The one to iterate with before a final pass.
  • Nano Banana LiteThe light variant of the previous generation, at 1K.

What Beemm Vision adds on top

Access to the model is not the point — Google offers it directly. The difference is what happens before and after the generation.

  • Compare without switching tabs: the same brief goes to Nano Banana, Seedream and FLUX side by side, and you keep the winner.
  • Chain: the result feeds an edit, crop or animation node in the same graph.
  • Replay: a nodal workflow reruns over a new list of products or references without rewriting a prompt.
  • Ship: a graph becomes a shareable app with its own input fields, for a team that will never open the canvas.
  • One credit balance covers the whole catalogue — no per-provider API keys to manage.

Nano Banana, Gemini, Pro, Lite: which is which

Nano Banana and Gemini are the same model

Nano Banana is the public nickname for Google's image models; inside Google they ship in the Gemini family. The name spread on public leaderboards before the official announcement, it stuck, and Google eventually adopted it. So "Google Nano Banana", "Gemini Nano Banana" and "Nano Banana" all point at one thing. On Beemm Vision you pick it by its short name in the model selector: no Google account, no API key, no Cloud project to open.

Nano Banana 2 or Nano Banana Pro

Both work the same way; they belong to different moments. Nano Banana 2, 0.5K to 4K, covers most of the work — and Nano Banana 2.1, the newest, keeps the same approach at 1K, 2K or 4K with up to fourteen reference images when editing. Nano Banana Pro is the quality tier: same brief, more finished render, 1K, 2K or 4K — the one you keep for the delivered version. The two Lite variants go the other way, at 1K: they are for finding the right image fast, before re-running the winner on Pro. You switch in the selector, mid-flow, without changing plan.

What about Nano Banana 3?

The model selector is the source of truth: it lists what the platform can actually run, not what has been announced. A new tier appears there the day it becomes executable, with nothing to change on your side — which is the point of a platform that is not tied to one vendor: when the next generation ships, it joins the catalogue instead of asking you to move.

Writing a prompt that holds on Nano Banana

Four habits are enough. Put the text you want rendered in quotes, literally: the model then treats the string as something to reproduce rather than a concept to interpret. Describe the medium before the content — "screen-printed poster", "35mm film photo", "matte 3D render" — because that frames everything else. Attach a reference image as soon as character consistency matters. And correct in plain language on the image you got, rather than rewriting the prompt from scratch: that is where this model is strongest.

Nano Banana makes images, not video

It is a still-image model, Pro variants included. To animate a visual, wire its output into a video node in the same graph — Kling or Seedance, for instance, both of which accept a starting image. The Nano Banana frame becomes the first frame, and the whole chain replays on the next batch in one click.

Frequently asked questions

Is Nano Banana free on Beemm Vision?
The free plan gives access to the whole catalogue, Nano Banana included, with 100 free credits. Each generation spends credits according to the model and settings you choose, and the cost is shown before you launch it.
What does Nano Banana 2.1 bring?
It is the newest version of the family, for text-to-image and editing alike. It outputs at 1K, 2K or 4K, takes up to fourteen reference images and offers fourteen formats, from an extra-wide 8:1 to an extra-tall 1:8.
What is the difference between Nano Banana and Nano Banana Pro?
Pro is the quality tier of the same family: a more finished render, with output up to 4K. The Lite variants go the other way — faster, at 1K, made for iterating before a final pass.
Can it write text in a language other than English?
Yes, and that is the model’s main strength. Accented characters and typographic marks are rendered correctly. Put the wanted text in quotes inside your prompt so the model treats it as a string to reproduce.
Who owns images generated with Nano Banana?
You keep the rights to what you generate, under the terms set out in our Terms of Service and in the licence of the model provider, which is named for every model in our public “Models and providers” annex.
Can I drive Nano Banana from Claude or Cursor?
Yes. The Beemm Vision MCP server exposes image generation to an agent: it reads the catalogue, picks the model, launches the generation and collects the result without leaving your editor.
Do I need a Google account to use Nano Banana?
No. The model is called from Beemm Vision with a single account — the platform one. There is no Google Cloud project to create, no API key to paste, and no separate bill to watch at the vendor's end.
Can Nano Banana generate video?
Not directly: Nano Banana produces still images. Inside a Beemm Vision workflow its output feeds a video node — Kling, Seedance or another animation model from the catalogue — which uses it as the first frame.
Is Nano Banana 3 available?
The catalogue exposes Nano Banana 2.1, Nano Banana 2, Nano Banana Pro and the two Lite variants. The model selector is the source of truth: it shows what the platform can run at the moment you open it, and a new tier shows up there as soon as it ships.

Try Nano Banana now

The button opens Generate with Nano Banana already selected: describe your image. The full catalogue is available from the free plan.

Open the studioSee the other modelsHow credits work

Keep reading