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

Nano Banana works best when you treat it as an editing loop: brief, generate, revise, inspect, export.

If you want to learn how to use Nano Banana, start with a small mindset shift.

Nano Banana is not just a text-to-image toy where you type a beautiful sentence and hope the model guesses what you mean. It is Google’s Gemini image generation and editing family, and the strongest results usually come from a simple loop: describe the job, protect the details that must not change, generate an image, then revise it with narrow follow-up instructions.

That matters because most Nano Banana failures are not mysterious. The product logo changes. A face drifts. A background edit alters the subject. Text looks almost right but not quite. The image is attractive, but it is not usable.

The fix is not a longer prompt. The fix is a more useful brief.

This guide explains how to use Nano Banana in the Gemini app, Google AI Studio, and the Gemini API. It covers the practical workflow, model selection, prompt structure, image editing, reference images, API usage, examples, troubleshooting, safety checks, and the final review process you should run before publishing an AI-generated image.

Quick Answer: How to Use Nano Banana

To use Nano Banana:

1. Open Gemini for casual image generation, Google AI Studio for prompt testing, or the Gemini API for product workflows.
2. Choose the right image model: Nano Banana 2 for general work, Nano Banana Pro for higher-stakes final assets, and Nano Banana 2 Lite for cheaper high-volume drafts.
3. Write a prompt that states the purpose, subject, style, references, output format, and constraints.
4. If editing an existing image, upload the image and say exactly what should change and what must stay the same.
5. Generate one result, inspect it, then revise with targeted follow-up prompts.
6. Before publishing, check identity, product details, text, logos, artifacts, rights, and whether the image should be labeled as AI-generated.

A good Nano Banana workflow is not one perfect prompt. It is a short loop of generation, narrow edits, and final review.

The shortest useful formula is:

Create a [type of image] for [purpose].
The subject is [main subject].
Use [reference image] to preserve [specific details].
Change only [the thing you want changed].
Output: [aspect ratio, style, framing, text rules].
Avoid [the failure modes that would ruin this image].

For example:

Create a 16:9 blog cover image for an article about AI image editing.
The subject is a clean desk with a laptop showing a before-and-after image edit.
Use the uploaded product image to preserve the product shape, color, and logo placement.
Change only the background to a bright studio setting.
Style: clean editorial tech, realistic lighting, no text in the image.
Avoid warped logos, unreadable UI text, extra products, and cyberpunk effects.

That is already much better than:

Make a cool AI image editing cover.

What Is Nano Banana?

Nano Banana is the nickname Google uses for Gemini’s native image generation and image editing models. In practical terms, it lets you create images from text, edit existing images with natural language, combine text and image inputs, and iterate conversationally.

As of September 15, 2026, Google’s Gemini API documentation lists a few Nano Banana image models:

Model nickname	Official model name	Best use
Nano Banana 2 Lite	gemini-3.1-flash-lite-image	Lower-cost drafts, high-volume generation, fast exploration
Nano Banana 2	gemini-3.1-flash-image	General image generation and editing
Nano Banana Pro	gemini-3-pro-image	Final assets, product visuals, ads, diagrams, and text-sensitive work
Nano Banana	gemini-2.5-flash-image	Older workflows that have not migrated yet

The names can be confusing because people often say “Nano Banana” when they mean the whole image model family. If you are using the Gemini app, you may not see the raw API model names. If you are using Google AI Studio or the API, model choice matters more.

The simple rule:

Use Nano Banana 2 for everyday image work. Use Nano Banana Pro when the result will be published, used in marketing, or judged by exact details. Use Nano Banana 2 Lite when speed and cost matter more than final polish.

Where Can You Use Nano Banana?

There are three common entry points.

Entry point	Best for	What it feels like
Gemini app	Casual creation and quick edits	Chat with Gemini, upload images, ask for changes
Google AI Studio	Prompt testing and model comparison	Choose models, test prompts, inspect outputs before coding
Gemini API	Apps, automation, batch workflows, developer products	Send text and image inputs, receive generated image data

If you are learning, start in the Gemini app or AI Studio. If you are building a repeatable workflow, move to the API after you know which prompts and models work.

Do not start with code if you have not tested the creative workflow. API code will not fix a vague prompt. It will only make vague prompts run faster.

Which Nano Banana Model Should You Choose?

Pick the model by job risk: draft cheaply, iterate with the balanced model, and finish important assets with Pro.

Model selection should follow the risk of the job.

Use Nano Banana 2 Lite for drafts and volume

Nano Banana 2 Lite is the economical choice. Use it when you need many rough directions, when you are testing prompt shapes, or when the image is not the final asset.

Good uses:

• Generating rough visual directions
• Trying 10 to 30 thumbnail ideas
• Drafting blog illustration concepts
• Low-risk internal mockups
• Batch exploration where cost matters

Do not use the cheapest model just because it is available. If the image includes product labels, brand details, people, or important text, test cheaply but finish with a stronger model.

Use Nano Banana 2 for everyday work

Nano Banana 2 is the default starting point for most people. It is the balanced choice for image generation, editing, prompt iteration, blog graphics, social concepts, and everyday creative work.

Good uses:

• Blog covers
• Social visuals
• Creative direction
• Background replacement
• Product concepting
• Lightweight diagrams
• Reference-image experiments

If you are not sure where to begin, start here.

Use Nano Banana Pro for final assets

Nano Banana Pro is the better choice when mistakes become expensive.

Use it for:

• Product images
• Ads
• Landing page hero images
• Brand campaigns
• Final client-facing assets
• Images with short text
• Diagrams where visual clarity matters
• High-trust edits to existing images

Even with Pro, do not skip review. Better models reduce errors. They do not remove the need for a human check.

Plan migration for older Nano Banana workflows

The older gemini-2.5-flash-image model still appears in official documentation, but Google describes it as the legacy Nano Banana model and recommends moving new work to the newer image models. If you built a workflow around the older model, check the current Google AI for Developers docs and migration notes before relying on it.

For new work, build on the current Gemini 3 image models unless you have a specific compatibility reason not to.

The Nano Banana Prompt Formula

Most bad prompts fail because they mix six different jobs into one vague sentence. A better prompt separates them.

Strong prompts separate the job, subject, references, edit target, output format, and constraints.

Use this structure:

Purpose: [where the image will be used and what it should do]
Subject: [main person, product, object, scene, or interface]
References: [how each uploaded image should be used]
Edit target: [what should change, if editing]
Output: [aspect ratio, framing, style, size, text rules]
Constraints: [what must stay unchanged and what to avoid]

Here is a complete prompt:

Purpose: Create a hero image for a blog post about AI product photography.
Subject: A skincare bottle on a clean bathroom counter with soft morning light.
Reference: Use the uploaded product image to preserve the bottle shape, cap color, label placement, and logo position.
Edit target: Change only the background from a cluttered room to a minimal white stone counter.
Output: 16:9 horizontal image, realistic editorial product photography, no text added.
Constraints: Keep the product proportions, label layout, and logo unchanged. Avoid extra bottles, misspelled text, warped packaging, plastic-looking reflections, and unrealistic shadows.

The important words are “preserve,” “change only,” and “avoid.”

Nano Banana can create impressive images from broad prompts, but production work depends on constraints. If you care about a logo, say so. If the face must stay the same, say so. If the background is the only thing that should change, say so.

How to Generate a New Image With Nano Banana

For a new image, you do not need to write a novel. You need to define the visual job clearly.

Start with:

• What the image is for
• What should be visible
• The style and mood
• The aspect ratio
• Whether text should appear
• What to avoid

Example:

Create a 16:9 editorial blog cover about AI image editing.
Show a designer's desk with a laptop displaying a simple before-and-after image comparison.
Style: clean modern tech editorial, realistic lighting, warm neutral desk, subtle blue and green accents.
Leave clean negative space on the right for a headline.
Do not include logos, readable brand names, human faces, robots, or text inside the image.

That prompt gives Nano Banana enough direction without overloading it.

If the first result is close, revise narrowly:

Keep the same composition and desk. Make the lighting brighter and remove the extra tablet on the left. Keep the laptop in the same position.

If the result is completely wrong, rewrite the original brief. Do not keep asking the model to fix a direction you do not want.

How to Edit an Existing Image With Nano Banana

Image editing is where Nano Banana becomes genuinely useful. The trick is to make the edit target narrow.

Weak edit prompt:

Make this product photo better.

Better:

Change only the background to a clean white studio setup.
Keep the product shape, color, label, logo placement, shadow direction, and camera angle unchanged.
Do not add text, extra products, hands, reflections, or decorative props.

For portraits:

Preserve the person's face, hairstyle, age, expression, clothing, and body proportions.
Change only the background to a bright home office with natural daylight.
Keep the skin texture natural. Avoid plastic skin, changed facial features, extra fingers, and distorted glasses.

For UI screenshots:

Use this screenshot as the layout reference.
Create a cleaner marketing mockup of the same interface.
Keep the sidebar, chart area, and card layout in the same positions.
Do not generate detailed fake text. Use simple blocks and short labels only.
Avoid extra buttons, warped browser chrome, and unreadable tiny text.

The smaller the edit, the easier it is to judge success. Background first. Then lighting. Then crop. Then final polish.

How to Use Reference Images

A reference image is not magic. You need to tell Nano Banana what role the reference should play.

A single uploaded image can mean many things:

• Preserve identity
• Preserve product shape
• Copy lighting
• Copy color palette
• Use as composition reference
• Use as texture reference
• Replace or edit only part of it

Be explicit.

Use Image 1 for product identity: preserve the bottle shape, label position, cap color, and front-facing angle.
Use Image 2 for lighting style only: match the soft morning light and white counter, but do not copy the objects.
Use Image 3 for composition only: product centered with negative space above.

This is especially important when you upload multiple references. If you do not assign roles, the model may blend them together in a way that looks nice but fails the actual job.

For product work, identity usually matters more than style. For moodboards, style matters more than exact objects. For UI images, layout matters more than tiny text.

Name that priority in the prompt.

How to Iterate Without Losing What Worked

Nano Banana is conversational, so use it conversationally.

Bad iteration:

Try again, make it better.

Good iteration:

Keep the same product, camera angle, and background. Make the image brighter, reduce the orange color cast, and remove the extra reflection under the bottle.

Use this revision structure:

Keep [what worked].
Change only [one or two specific issues].
Avoid [the failure seen in the last result].

Examples:

Keep the same portrait and outfit. Change only the background from office to outdoor cafe. Avoid changing the face or hairstyle.

Keep the same composition. Make the laptop screen simpler and remove the fake text. Use clean blocks instead of readable labels.

Keep the product centered and fully visible. Make the shadow softer and more realistic. Do not change the label.

Do not change everything in one follow-up. If you ask for a new background, new lighting, new crop, new style, and new object placement, you are no longer editing. You are regenerating.

Prompt Templates You Can Copy

Blog cover

Create a 16:9 blog cover image for an article about [topic].
Show [main visual scene].
Style: [editorial, clean SaaS, realistic product photography, hand-drawn diagram, etc.].
Leave [left/right/top] negative space for a headline.
Do not include text, logos, watermarks, or human faces unless requested.
Avoid generic AI brain imagery, unreadable UI text, and visual clutter.

Product background replacement

Edit the uploaded product image.
Change only the background to [new background].
Keep the product shape, color, material, logo, label placement, camera angle, and shadow direction unchanged.
Output: [aspect ratio].
Avoid extra products, fake labels, warped packaging, changed proportions, and unrealistic reflections.

Portrait background edit

Edit the uploaded portrait.
Preserve the person's face, hairstyle, age, expression, clothing, and body proportions.
Change only the background to [new setting].
Lighting should look natural and match the subject.
Avoid identity drift, plastic skin, changed clothing, extra hands, and distorted glasses.

Social media image

Create a [1:1 / 4:5 / 9:16] image for [platform].
The image should communicate [one clear idea].
Subject: [person, product, object, scene].
Style: [creator-style, editorial, clean product, playful illustration].
Keep the composition simple and readable on a phone.
Avoid small text, clutter, fake brand marks, and busy backgrounds.

Diagram or infographic

Create a clean educational diagram explaining [concept].
Use [number] labeled sections: [labels].
Use simple shapes and large readable labels.
Style: light editorial infographic, no brand logos.
Avoid long paragraphs, tiny text, fake data, and decorative clutter.

For diagrams with important text, I still prefer deterministic SVG or a design tool. Image models are improving, but exact text remains one of the first things to check.

How to Use Nano Banana in the Gemini API

If you are building an app, use the Gemini API rather than manually downloading images from the Gemini app.

The concept is simple:

1. Create a Gemini API client.
2. Choose an image-capable model.
3. Send a text prompt, optionally with image inputs.
4. Ask for image output.
5. Save the returned image bytes.
6. Store the prompt, model name, input references, and output metadata for repeatability.

Google’s current Python SDK examples use client.interactions.create and response_format to request image output. A simplified version looks like this:

from google import genai
import base64

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-3.1-flash-image",
    input="Create a clean 16:9 blog cover about AI image editing. No text.",
    response_format={
        "type": "image",
        "aspect_ratio": "16:9",
        "image_size": "2K",
    },
)

if interaction.output_image:
    image_bytes = base64.b64decode(interaction.output_image.data)
    with open("nano-banana-cover.png", "wb") as f:
        f.write(image_bytes)

Treat this as a conceptual pattern, not a frozen snippet. SDK names, helper methods, and model availability can change, so check the current Gemini API docs before implementing production code.

For production workflows, log:

• Prompt
• Model name
• Date and time
• Input image IDs
• Output file path
• Aspect ratio
• Safety or policy blocks
• User-facing revision history
• Cost and latency

Image generation is probabilistic. Logs are the difference between an experiment and a workflow you can debug.

What Nano Banana Is Good At

Nano Banana is strongest when you give it a clear visual job.

Good use cases:

• Blog and article images
• Product concept photos
• Background replacement
• Social media visuals
• Moodboards
• Marketing mockups
• UI concept images
• Simple diagrams
• Character or portrait edits with clear preservation instructions
• AI art refinement
• Internal creative exploration

It is especially useful when you need to move fast from idea to image without opening a full design stack.

The sweet spot is not “make anything.” It is “make this specific image usable for this specific purpose.”

What to Be Careful With

Some jobs still need extra caution.

Text inside images

Short labels may work. Long paragraphs, legal copy, product claims, and exact typography are still risky.

For important text, generate the visual without text and add copy later in a deterministic editor.

Use:

Leave clean negative space for a headline. Do not render the headline inside the image.

Logos and product labels

AI models can distort logos, move labels, or invent extra brand marks. If a logo matters, preserve it explicitly and inspect the result at full size.

Faces and identity

For portraits, repeat the preservation instruction in every edit. Check face shape, eye direction, hair, age, skin texture, hands, and clothing.

Real people, brands, and rights

Do not upload or transform images you do not have the right to use. Be careful with celebrities, private people, children, trademarks, copyrighted characters, and confidential client assets.

AI editing can make an image more publishable visually while making it more risky legally or ethically.

Misleading edits

If an AI-edited image could mislead viewers about a real person, product, location, or event, label it clearly or do not use it. Google also uses SynthID and watermarking approaches across its AI image ecosystem, but watermarks are not a substitute for honest publishing context.

A Practical Review Checklist

Before you publish a Nano Banana image, check:

• Does the subject still match the intended person, product, or object?
• Did the model change any logo, label, packaging, or interface detail?
• Is all visible text accurate and readable?
• Are hands, faces, shadows, reflections, and edges believable?
• Does the final image match the requested aspect ratio?
• Is the crop usable on the target platform?
• Is the file size appropriate for the web?
• Do you have rights to the source images and final use?
• Should the image be disclosed as AI-generated or AI-edited?
• Would a viewer reasonably misunderstand the image as a real photo?

If the image is for ecommerce, check the product twice. A beautiful image that changes the product is a bad product image.

If the image is for news, health, finance, law, politics, or public figures, raise the review bar. Synthetic visuals in high-trust contexts need more care.

Common Nano Banana Mistakes

Mistake 1: Prompting for style before purpose

Weak:

Cinematic, beautiful, premium, ultra realistic.

Better:

Create a 16:9 hero image for a SaaS landing page showing a calm analytics dashboard on a laptop in a bright workspace.

Style helps. Purpose leads.

Mistake 2: Not saying what must stay unchanged

If editing a product, portrait, or UI, the prompt should include preservation rules. Otherwise the model may improve the image by changing the exact thing you needed to keep.

Mistake 3: Asking for too many edits at once

Change the background first. Then lighting. Then crop. Then details.

One controlled change beats five vague changes.

Mistake 4: Trusting image text

Do not assume generated text is correct. Zoom in. For ads, product labels, legal disclaimers, charts, and UI, add final text manually when precision matters.

Mistake 5: Using Pro for every draft

If you need exploration, start cheaper. Generate broad directions with Nano Banana 2 Lite or Nano Banana 2, then use Nano Banana Pro for the final pass.

Mistake 6: Skipping rights review

Upscaling, editing, restyling, or regenerating an image does not magically clear rights. The original input still matters.

Example Workflows

Workflow 1: Blog cover

1. Write a plain description of the article’s core idea.
2. Ask for a 16:9 editorial image with no text.
3. Generate three to six directions.
4. Pick one.
5. Use follow-up prompts to adjust lighting, crop, and clutter.
6. Add title text later in your CMS or design tool if needed.

Prompt:

Create a 16:9 editorial blog cover for an article about learning to use Nano Banana.
Show a clean AI image workflow on a laptop: prompt panel, image preview, and review checklist.
Style: bright modern tech editorial, realistic desk, soft daylight, muted blue and amber accents.
No text, no logos, no people, no robots, no watermark.

Workflow 2: Product photo cleanup

1. Upload the product photo.
2. Ask for a background-only change.
3. Preserve product identity aggressively.
4. Review label, color, proportions, and shadows.
5. Export only after comparing with the original.

Prompt:

Edit this product photo.
Change only the background to a clean white studio surface with soft natural shadow.
Keep the product shape, color, label, logo placement, camera angle, and size unchanged.
Avoid extra labels, changed packaging, unrealistic reflections, and distorted text.

Workflow 3: Social media creative

1. Define the platform ratio first.
2. Use one idea per image.
3. Keep mobile readability in mind.
4. Avoid tiny text inside the image.

Prompt:

Create a 4:5 vertical social image about AI image editing.
Show a simple before-and-after photo edit on a phone screen, with a clean desk background.
Style: bright creator-style image, realistic but polished.
Keep the composition simple and readable on mobile.
Do not include readable text, logos, or watermarks.

Workflow 4: API-powered image feature

1. Prototype prompts in AI Studio.
2. Select the model based on cost, quality, and latency.
3. Build an API call with image output enabled.
4. Save input and output metadata.
5. Add human review for public-facing assets.
6. Add fallback behavior for blocked or failed generations.

This is where many teams go wrong. They code first, then discover the creative workflow is inconsistent. Test the image behavior before you build the interface around it.

FAQ

What is Nano Banana?

Nano Banana is Google’s nickname for Gemini’s native image generation and editing models. It can create images from text, edit uploaded images, combine text and image inputs, and support conversational revisions.

How do I use Nano Banana?

Use the Gemini app for quick image generation and editing, Google AI Studio for prompt testing, or the Gemini API for developer workflows. Write prompts that define the purpose, subject, references, edit target, output format, and constraints.

Is Nano Banana free?

Availability, free access, and limits depend on the product surface, account, region, and current Google pricing. Check the live Gemini app, Google AI Studio, and Gemini API pricing pages before building a workflow around a specific free limit.

What is the best Nano Banana model?

For most users, Nano Banana 2 is the best starting point. Nano Banana Pro is better for final product, marketing, and text-sensitive images. Nano Banana 2 Lite is better for lower-cost drafts and high-volume exploration.

Can Nano Banana edit existing images?

Yes. Upload the image and tell it exactly what to change and what to preserve. Use phrases like “change only the background” and “keep the product shape, label, color, and logo unchanged.”

Can Nano Banana create images with text?

It can create images that include text, and newer Gemini image models are better at this than older systems. Still, you should inspect all text. For important headlines, legal copy, logos, labels, and UI text, add the final text in a design tool.

Can I use Nano Banana for commercial images?

Commercial use depends on the Google product terms, your account, your input images, and the final use case. You still need rights to the images you upload and a safe reason to use the generated result.

Does Nano Banana add watermarks?

Google’s public materials describe SynthID and watermarking for AI-generated images in its Gemini image ecosystem. The exact user-visible behavior can depend on product surface and release stage, so check the current Gemini documentation for the tool you are using.

What is the best prompt for Nano Banana?

The best prompt is not a magic sentence. Use a structured brief: purpose, subject, references, edit target, output format, and constraints. The more important the asset, the more specific you should be about what must not change.

Should I use Gemini app, AI Studio, or API?

Use Gemini app for quick personal creation, AI Studio for testing prompts and comparing models, and the API when you need automation, repeatability, logging, or a product feature.

Final Verdict

The best way to use Nano Banana is to treat it like a visual editor, not a slot machine.

Start with the job. Name the subject. Tell each reference image what role it plays. Make one change at a time. Preserve the details that matter. Then inspect the result like someone who has to publish it, not like someone impressed by the first pretty draft.

That workflow is slower than typing “make it cinematic.”

It is also the difference between an image that looks interesting and an image you can actually use.

Sources Checked

This article was prepared with public product information checked on September 15, 2026:

• Google AI for Developers image generation documentation: https://ai.google.dev/gemini-api/docs/image-generation
• Google AI for Developers image generation models: https://ai.google.dev/gemini-api/docs/models
• Google AI for Developers pricing: https://ai.google.dev/gemini-api/docs/pricing
• Google Gemini app image editing announcement: https://blog.google/products-and-platforms/products/gemini/updated-image-editing-model/
• Google DeepMind SynthID: https://deepmind.google/models/synthid/