The field guide
Nano Banana
A practical field guide to Google's image models. Understand the names, make your first image, and learn how to ask for the edit you actually want.

What is Nano Banana?
Nano Banana is the name Google uses for Gemini's native image-generation capabilities. You describe a picture to create, or upload a picture and describe a change. The model produces a new image in response.
The name now covers a family of models, not one app or a single subscription. That distinction matters: the image model, the Gemini app, and a third-party website can have different features, prices, and limits.
One name. A few different models.
Google's image-generation documentation lists the following models as of September 6, 2026. The names in an app's model menu do not always match the image model used underneath.
| Name | Google model | How Google positions it |
|---|---|---|
| Nano Banana 2 Lite | Gemini 3.1 Flash Lite Image | Fast, lower-cost image generation. Not the first choice for complex sequential editing. |
| Nano Banana 2 | Gemini 3.1 Flash Image | A balance of speed, image quality, and instruction-following. |
| Nano Banana Pro | Gemini 3 Pro Image | More demanding compositions and detailed creative control. |
| Original Nano Banana | Gemini 2.5 Flash Image | The earlier model. Google now describes it as a legacy option. |
You do not need an API key to follow the Gemini app tutorial. If you see an API model ID in an older guide, check the current Google documentation before using it in code.
What would you like to do?
- Make your first image
A step-by-step guide to opening image creation, describing a scene, editing a photo, and saving the result.
- Find a better starting prompt
Original examples for portraits, backgrounds, readable text, and specific edits, with instructions for what to preserve.
- Understand free access and Pro
What is included in Gemini, what changes with a paid plan, and why third-party credits and API bills are separate.
Before your first image
A useful instruction is concrete: “Change the wall to a muted green; keep the furniture, window, and camera angle unchanged.” A less useful instruction is “Make it amazing.” Start small, compare the result with your intention, and revise one thing at a time.
Check faces, logos, exact wording, and small details before using an image. Generative editing is not the same as selecting a layer in a conventional photo editor. A request to preserve something is an instruction, not a guarantee.
Use images you have permission to upload. Consider what an image reveals about other people before sending it to any AI service, and check the provider's privacy and usage terms. Google states that generated images include a SynthID watermark.
This is an independent guide. For model access, use the official Gemini app. A site with “Nano Banana” in its domain is not automatically operated by Google.
FAQ
Common questions
Is Nano Banana a separate app?
Nano Banana is a name used for Google's image-model family. You can use supported models inside Gemini and through developer services. A website with Nano Banana in its name is not necessarily operated by Google.
Can it edit an existing photo?
Yes. Supported image models can work with an uploaded image and a text instruction. Specify the change and the details that must stay the same; inspect the result because identity, text, and small details can still change.
Is this the official Nano Banana website?
No. Nano Banana For Photos is an independent educational website. Official Google pages are linked throughout the guides; this site does not provide a Google subscription or image-generation service.
Can Nano Banana generate video?
The models discussed here generate and edit images. Video generation is a separate workflow and may use a different model or service. A third-party video feature is not evidence that an image model natively generates video.
Sources & editorial notes
Access information and model names were checked against the official sources below. Interfaces, limits, and offers can change.
Independent editorial writing. Original prompt examples are not a model benchmark; external demonstrations carry their own attribution.
Keep exploring