Nano Banana 2, 2.1, Lite and Pro: Which Model Is Which

Google's image models have gathered a confusing set of names: nicknames, version numbers and the formal Gemini model names, sometimes all for the same thing. The confusion matters because the models differ in resolution, how many reference images they take, whether they can search the web and how much they cost. This guide lines them up, using Google's Gemini API documentation and the Gemini app's own model menu, so you know which one you are using and which to ask for.

The line-up

Google also notes that its older Imagen text-to-image models are legacy, and that every generated image carries an invisible SynthID watermark.

In the Gemini app

Gemini's overview page describes Nano Banana 2 as the free image generation and editing model, and Nano Banana 2.1 as "our best image generation and editing model", available with a Google AI plan. To use them, open the tools menu and choose Create images, then pick Fast, Thinking or Pro in the model menu, and type a prompt or upload a photo to edit. Google AI Plus, Pro and Ultra subscribers can also rerun an image with Nano Banana Pro from the three-dot menu, using Redo with Pro. Paid plans have higher generation limits.

In Google Flow

Flow uses the same family for frames and ingredients: Nano Banana Pro (the default for Ultra subscribers), Nano Banana 2.1, and Nano Banana 2 Lite, which Flow's help describes as available at no charge. See Flow frames to video for how those images feed video.

The capabilities that differ

• Resolution. The Gemini 3 image models generate 1K by default and can output 2K and 4K; Nano Banana 2 also offers a small 512-pixel size; Nano Banana 2 Lite is 1K only.
• Reference images. Up to 14 in total. Nano Banana 2.1 and Nano Banana 2 take up to 10 high-fidelity object images and up to 4 character images; Nano Banana Pro up to 6 object images, 5 character images and 3 style references; 2 Lite up to 14 object images.
• Search grounding. All except 2 Lite can use Google Search to ground images in current facts such as weather or recent events; 2.1 and Nano Banana 2 can also use Google Image Search as visual context.
• Thinking. The Gemini 3 image models reason through complex prompts and may test the composition with interim images; in the API, 2.1 lets you choose minimal, medium or high thinking.
• Aspect ratios. Common ratios from 1:1 to 21:9, including 4:5, 3:2 and 9:16.
• Video input. 2.1, Nano Banana 2 and 2 Lite can take a video as a reference for a new image, for thumbnails or posters.

Which one to use

For prompts that suit these models, see the Nano Banana prompt guide.

Using the models outside Google

The same Nano Banana models appear in other platforms as hosted options, including Leonardo and Artlist at the time of writing, sometimes under slightly different names or resolution choices. The model's behaviour is the same, but each platform wraps it in its own settings, limits and credit costs, and may not expose features such as search grounding or thinking levels. If a result differs between the Gemini app and another tool, check which Nano Banana version and resolution each one used before blaming the prompt.

Keeping track of versions

Because the same image can come from different Nano Banana models depending on your plan, menu choice and limits, record which model made each image you keep. Google's overview page also notes that limits apply; on some plans you may fall back to a lighter model once a limit is reached, which changes the look. A note of the model, the date and the prompt makes a later revision far easier.

Finally, remember that names will keep changing. When Google releases a new version, the old one usually becomes legacy rather than disappearing at once, so older guides and saved prompts keep working for a while.

Try it

A clean, modern infographic titled "How rain forms", three labelled stages from left to right: "Evaporation", "Condensation", "Precipitation". Soft blue and white palette, simple flat icons, sans-serif type, no other text, 3:2 format.

Generate it in the Gemini app with Fast, then with Thinking, then (if your plan allows) Redo with Pro. Compare the spelling of the labels and the layout across the three. That one test shows more about the models than any table.