How to Use Nano Banana: A Practical Guide to Image Generation and Editing

The first thing that trips people up with Nano Banana is not prompt grammar.

It is failing to separate what should change from what must stay the same.

Nano Banana is useful because it can generate images from text, edit uploaded images, and refine a result through conversation. That also means it can drift if the instruction is vague. A face may change, a product shape may shift, text may become slightly wrong, or the model may create something that only feels close to your idea.

This guide focuses on a workflow that reduces those failures.

What Nano Banana is

Nano Banana is the image generation and editing family available through Gemini. In the Gemini API, the practical model names are Nano Banana 2 Lite, Nano Banana 2, Nano Banana Pro, and the earlier Nano Banana model.

Name	Model ID	Best use
Nano Banana 2 Lite	gemini-3.1-flash-lite-image	Low cost, low latency, high-volume drafts
Nano Banana 2	gemini-3.1-flash-image	Everyday generation and editing
Nano Banana Pro	gemini-3-pro-image	Product images, ads, diagrams, text-heavy assets
Nano Banana	gemini-2.5-flash-image	Stable older workflow and migration cases

If you are unsure, start with Nano Banana 2. If the image will be used publicly in ads, ecommerce, landing pages, or client work, use Nano Banana Pro for the final pass.

Three ways to start

Entry point	Best for	What you can do
Nano Banana AI Studio	Fast browser workflow	Generate and edit Nano Banana images directly on this website
Gemini app	Quick personal testing	Text-to-image and simple uploaded-image edits
Google AI Studio	Prompt testing	Compare models and outputs before using the API
Gemini API	Products and operations	Automate generation, batches, multi-image inputs, and conversational editing

For the fastest web workflow, start with Nano Banana AI Studio on this website. The Gemini app is enough for casual use, Google AI Studio is useful for prompt and model testing, and the API is the right option when image generation becomes part of a product, SaaS workflow, ecommerce pipeline, or internal content process.

Basic workflow

A stable beginner workflow looks like this:

1. Open Nano Banana AI Studio for the fastest web workflow. You can also test in Gemini or Google AI Studio.
2. Choose a Nano Banana image model.
3. Describe the image in plain language.
4. Upload a reference image if needed.
5. Review the result and revise it conversationally.
6. Check text, faces, hands, logos, rights, and misuse risk before publishing.

Do not make the first prompt too crowded. Start with the structure.

Example:

Tokyo cafe, a freelancer using a laptop to edit AI images, natural light, calm mood, 16:9.

Then refine:

Keep the person and background the same. Show a before and after product image on the laptop screen.

Nano Banana usually works better as an editing conversation than as a one-shot poster machine.

Prompt in four parts

Use four blocks: purpose, subject, constraints, and output.

Part	What to write	Example
Purpose	Where the image will be used	Blog cover, ad banner, product photo
Subject	What appears in the image	Person, product, UI, landscape, diagram
Constraints	What must not change	Keep face, change only background, no logo
Output	Format and style requirements	16:9, 1:1, no text, extra whitespace

Example prompt:

Purpose: blog cover. Subject: a workspace where an AI image editing tool is improving a product photo. Constraints: keep the product shape and color unchanged; change only the background. Output: clean white studio style, no text, 16:9.

The word only matters. Background only. Color only. Composition only. A focused edit is easier to control than a broad transformation.

Image editing tips

Nano Banana is strongest when it edits existing images.

For a product photo, write:

Change only the background to a bright white studio. Keep the product shape, logo, color, and shadow direction unchanged.

For a portrait, write:

Keep the face, hair, age, and clothing unchanged. Change only the background to a minimal office with natural light. Do not over-smooth the skin.

For a UI screenshot, write:

Keep the dashboard layout unchanged. Change only the color palette to dark mode. Do not move charts, tables, or buttons. Do not add new text.

Small edits are easier to verify. Change background, then lighting, then color. Do not ask for every change at once.

Nano Banana 2 vs Pro

Nano Banana 2 is for exploration. Use it to generate many directions cheaply and quickly.

Nano Banana Pro is for final assets. Use it when the image carries trust: ads, product shots, diagrams, brand creatives, localized visuals, and images with visible text.

Situation	Recommended model
Blog drafts	Nano Banana 2
Social mood images	Nano Banana 2
Ecommerce final product photo	Nano Banana Pro
Advertising banner	Nano Banana Pro
Text inside the image	Nano Banana Pro
API batch generation	Nano Banana 2 Lite or Nano Banana 2

Text inside images still needs human review, especially long Japanese or English copy. Even strong text rendering can produce small mistakes.

Practical prompts

Blog cover:

Create a 16:9 blog cover about using an AI image editing tool. A laptop screen shows before and after image editing. No human face. Calm technology mood. No text.

Product background change:

Change only the background of this product image. Keep the product, label, logo, color, and shape unchanged. Use a bright white studio with a natural shadow.

Portrait lighting:

Keep the face, age, hairstyle, and clothing unchanged. Make only the lighting look like natural daytime light. Do not over-smooth skin. Clean up the background slightly.

Diagram:

Create a simple diagram explaining the Nano Banana workflow: input, generation, editing, review. Use wide whitespace and a 16:10 blog-friendly layout.

API use

Developers can call the Gemini API with an image-capable model and request image output. The exact SDK details may change, so implementation should follow the current official documentation. In production, the important part is logging: model ID, prompt, input image, ratio, output format, time, cost, and failure reason.

Without logs, image generation becomes guesswork. With logs, you can reproduce results, compare models, and understand why a workflow changed.

Common failures

Most failures come from unclear instructions rather than a bad model:

• The prompt is vague.
• The edit changes too many things at once.
• The prompt does not say what to preserve.
• Text inside the image is not checked.
• The user has not confirmed rights for reference images.

Be especially careful with real people, real products, real places, and news-like imagery. Make clear when an image is AI generated. Do not create misleading edits. For higher-risk use, consider watermark and provenance checks such as SynthID.

Recommended workflow

1. Generate 6 to 12 directions with Nano Banana 2.
2. Choose one direction.
3. Edit background, composition, and color in separate steps.
4. Use Nano Banana Pro only for the final version.
5. Review text, faces, hands, product logos, rights, and possible misunderstanding.

Image generation is not a lottery ticket. It is an editing process.

FAQ

Can I use Nano Banana for free?

Free access depends on the Google product, account, region, model availability, and current limits. For production API use, check the current Google AI for Developers pricing page.

Can I prompt in Japanese?

Yes. Japanese prompts work well for detailed instructions. If the image contains Japanese text, review every character before publishing.

Can I change only the background without changing a face?

Yes, but say it directly. Write that the face, hair, clothing, and age must stay unchanged and that only the background should change.

Which model should I choose?

Use Nano Banana 2 for drafts, blogs, and social images. Use Nano Banana Pro for ads, ecommerce, final visuals, and images with text.

Can I use the output commercially?

Commercial use depends on the Google service, contract, input image rights, and the final use case. Be careful with people, trademarks, existing characters, and third-party photos.

Summary

Using Nano Banana well means deciding what to create, what not to change, generating small drafts, editing through conversation, and reviewing the final image manually.

If you hand everything to the model, the result often feels generic. If you act like an editor, Nano Banana becomes much more useful.

References

• Google AI for Developers Nano Banana image generation
• Gemini 3 Pro Image model page
• Gemini 2.5 Flash Image model page
• Gemini Developer API pricing
• Google DeepMind SynthID