How to write Nano Banana prompts for Google's Gemini image model: subject, style, framing, light, text, editing, 18 marketing examples and image SEO tips.

In short
Nano Banana is the nickname of Google's image generation and editing model in Gemini, first released as Gemini 2.5 Flash Image in August 2025, followed by Nano Banana Pro (Gemini 3 Pro Image) in November 2025 and a newer version in 2026. A good Nano Banana prompt describes the subject, the setting, the style, the framing, the light and any text to display, in full sentences. For edits, describe only what must change and what must stay. Images carry an invisible SynthID watermark, and once published they need the usual image SEO: descriptive file name, alt text, light WebP files.
Want to go further? Ask
Nano Banana prompts are the instructions you write to create or edit an image with this model, in the Gemini app or through the Gemini API, which Google documents in its image generation guide. The model is multimodal: it understands text and images together, which makes it strong at editing a photo you upload and at keeping a character or a product consistent across several images.
The name started as an internal codename that spread among users before Google adopted it publicly. Behind it are Google's image models in Gemini:
| Name | Model | Release | Strengths |
|---|---|---|---|
| Nano Banana | Gemini 2.5 Flash Image | August 2025 | Fast generation, natural-language editing, character consistency |
| Nano Banana Pro | Gemini 3 Pro Image | November 2025 | Higher resolution, better text rendering, complex compositions |
| Newer versions | Later Gemini image models | 2026 | Faster and cheaper generation, check Google's documentation for current names |
Model names and limits change quickly: check the Gemini app or the API documentation for the version available to you.
The model responds best to descriptive sentences, the way you would brief a photographer or an illustrator, rather than to lists of keywords. Cover these elements:
For editing an existing image, upload it and write what changes and what stays: "Replace the background with a white studio wall, keep the product, its shadow and its colors unchanged." The general method of writing prompts is the same as for text; see our guide to AI prompts.
A generated image is a normal image for search engines. Google's guidance on image SEO best practices applies in full:
| Step | Good practice | Why |
|---|---|---|
| File name | Descriptive words: black-ceramic-coffee-mug.webp | Helps Google understand the image |
| Alt text | One factual sentence describing the image | Accessibility and context, see alt text |
| Format | WebP or AVIF, right dimensions | Faster pages |
| Weight | Compress before upload | Core Web Vitals |
| Placement | Near the text it illustrates, with a caption if useful | Relevance of the page |
| Structured data | Image in Product or Article markup | Rich results |
A free image compressor reduces the weight of generated files, which are often large. Our glossary entry on image optimization covers the rest of the checklist.
Nano Banana works best with prompts written like a creative brief: a precise subject, a setting, a style, a framing, a light and the exact text, plus clear instructions on what to keep when editing. Images then need the same care as any other visual: verification, honest use and image SEO, so that they help your pages rank instead of only filling space.
Google offers access to its image model in the Gemini app, with usage limits that depend on the account and the subscription. Developers use it through the Gemini API, which is billed by usage. Limits and plans change regularly, so check the Gemini app or Google's API pricing page for current conditions.
Google's terms generally allow you to use the images you generate, but you remain responsible for respecting the rights of others, such as trademarks, likenesses and copyrighted works, and for complying with advertising rules. Read the current terms of the service you use before publishing images in ads or on product pages.
Yes. Images created or edited with Google's image models carry SynthID, an invisible digital watermark designed to identify AI-generated content, and Google provides tools to check for it. Detection does not by itself harm SEO: what matters for ranking is that the image is relevant, honest and well optimized.
THEY TRIED SORANK BEFORE YOU
