How to write prompts for Nano Banana 2: Google's official image generation guide

Google has released an official guide on writing prompts for Nano Banana 2, the updated image generation model based on Gemini 3.1 Flash. The document contains specific techniques that help control style, composition, and detail. Here are the most useful tips.
Five elements of an effective prompt
Google recommends structuring your prompt into five blocks. The more detail you provide, the more accurate the result.
- Style: Illustration, photograph, watercolor, retro-futurism, abstraction — specify the genre clearly.
- Subject: Who or what is in the frame, appearance, clothing, pose. For example: a person in a blue jacket sitting at a desk.
- Setting: Where the action takes place — a cafe on Mars, an alchemist's library, a meadow at sunset.
- Action: What is happening — a character brewing coffee, petting a cat, reading a book.
- Composition: Angle and format — portrait, landscape, close-up, low-angle shot, wide-angle.
An example prompt from the guide: A baker's guide to different types of flour. A rustic, hand-drawn style comparing the protein content, best uses, and texture of each flour, displayed as sacks of flour. The model creates an infographic with flour sacks, labels, and a table of characteristics.
How to edit images via prompts
Nano Banana 2 allows you to refine images that have already been generated. Google suggests five ways to iterate:
- Character replacement: Change the subject while keeping the style and environment.
- Composition change: Ask the model to show the scene from a different angle — it will add details visible only from the new perspective.
- Action change: Have the character strike a different pose or perform a different action.
- Environment overhaul: Replace the background entirely or swap out individual elements like textures or objects.
- Style revision: Turn a photograph into an illustration, a cartoon drawing, or a Cubist-style painting.
The main advantage is that you don't need to rewrite the entire prompt. Just specify what needs to be changed, and the model will preserve the rest.
Three techniques for precise results
Google highlights several techniques that work particularly well:
Generating text in images. Enclose the desired words in quotation marks and describe the font. For example: Create a poster with the text "Wave Form" in bold sans-serif font. The model will correctly render the text with the specified typography.
Using real-world knowledge. Nano Banana 2 connects to the Gemini knowledge base and web search. You can request an infographic about the layers of the Earth or a water cycle diagram — the model will pull up current data and visualize it.
Generating multiple variations. Ask the model to create three variations of a product layout or four color palettes. This speeds up the selection process, so you don't have to generate images one by one.
Why this matters
Structured prompting saves time. Instead of dozens of attempts, you get the desired result on the first or second try. This is especially useful for those creating marketing materials, presentations, or educational content.
Nano Banana 2 is available in Google AI Studio and via API. The model works faster than its predecessor and follows instructions more accurately, especially when rendering text and complex compositions.
Want to try Nano Banana 2 and other Gemini models right now? GEMERA AI provides access to GPT, Claude, and Gemini for generating text, images, video, and music. Everything is available in one Telegram bot and web dashboard.