How to Use Nano Banana: Generate, Edit and Blend Images

Here is how to use Nano Banana: open the Gemini app, choose Images from the sidebar and describe the image you want. To edit a photo, upload it and tell Gemini what to change in plain words. That is the whole entry point, and it is free to start with a Google account.

Everything past that is technique. Most people who give up on Nano Banana after a few tries made the same three mistakes: a list of keywords instead of a sentence, five edits crammed into one message, and a square image cropped to 16:9 afterwards. The model is rarely the problem. This guide covers where to find it, how to generate and edit, how to blend several images into one, and how to use it for YouTube scene images and thumbnails without placing every picture by hand.

Which Nano Banana are you actually using?

When you open Gemini today, you are using Nano Banana 2, Google’s name for Gemini 3.1 Flash Image. Google released it on February 26, 2026, and it replaced earlier models as the default image engine across Gemini’s Fast, Thinking and Pro modes.

The name covers a family, which is why guides disagree:

Name	Official model	Released
Nano Banana	Gemini 2.5 Flash Image	August 26, 2025
Nano Banana 2	Gemini 3.1 Flash Image	February 26, 2026
Nano Banana 2 Lite	Gemini 3.1 Flash-Lite Image	June 30, 2026

All of them come from Google DeepMind. The original went viral after it appeared anonymously on a blind-test leaderboard in August 2025, and Google’s developer docs now list it as the legacy model. You don’t pick between them in the Gemini app; Gemini routes image requests to Nano Banana 2, or to the Lite version when you chat in Flash-Lite mode. Nano Banana Pro is a separate, heavier model (Gemini 3 Pro Image), and paid Google AI subscribers can rerun a finished image through it from the More menu with Redo with Pro.

Where can you use Nano Banana?

Nano Banana 2 runs in more places than most people realise. Pick the one that matches the job.

• The Gemini app, on the web at gemini.google.com and on Android and iOS. It is the easiest route, and the one most of this guide uses.
• Google AI Studio (aistudio.google.com), which exposes controls the app hides, including aspect ratio and 4K output.
• Google Search. AI Mode and Google Lens both run Nano Banana 2, so you can create or edit an image without leaving search.
• Flow, Google’s filmmaking tool, where Nano Banana 2 is the default image model.
• The Gemini API and Vertex AI, for developers and teams building it into their own software. The API model ID is gemini-3.1-flash-image.
• Third-party apps that license the model. TubeGen is one, using Nano Banana 2 to generate scene images for YouTube videos.

How to use Nano Banana to generate an image

In the Gemini app, generation takes four steps.

1. Go to gemini.google.com (or open the mobile app) and sign in.
2. Open the sidebar and click Images.
3. Pick a template, or type a description of the image you want.
4. Send it, then reply in the same chat to adjust what came back.

Step four is where the model earns its reputation. Nano Banana is conversational, so you refine the image rather than starting over. “Make it dusk,” “move the camera lower,” “drop the second car” all work as follow-ups, and the rest of the image stays put.

Google’s own guidance for the first request is sensible: lead with an action word like create or draw, name the style, then describe the subject, what it is doing, and where. A full sentence beats a pile of keywords. “A lighthouse keeper climbing a spiral staircase at night, lit by a single lantern, painted in thick oil strokes” gives the model a scene. “Lighthouse, night, oil painting, dramatic” gives it a guessing game.

Gemini doesn’t show an aspect ratio button in every view, so say the shape you want in the request. Write “wide 16:9 image” for a YouTube frame or “vertical 9:16” for Shorts, and check the result before you build on it.

How do you edit an existing photo with Nano Banana?

Upload the photo, click Chat, and describe the change. That is the core of it, and it is the feature that made the original model famous.

The full path on desktop:

1. Open gemini.google.com and click Library in the sidebar, or attach a new photo in any chat.
2. Click the image you want to change, then click Chat under it.
3. Describe the edit and submit.

What it handles well: replacing a background, removing an object or person, adding an element, changing the time of day, restyling a photo into an illustration, and translating text that already appears in the image. Google’s model page flags masked edits and major lighting changes as the weaker spots, so expect a couple of retries there.

Two habits make editing far more reliable. Change one thing per message, because stacking five edits into one request is how faces drift and details go missing. And name what must not change: “keep his face and jacket exactly the same, replace only the background with a rainy street” protects the parts you care about.

Editing uploaded images is limited to users 18 and over. If the edit option isn’t there, check the account’s age settings before anything else.

How do you blend images and keep a character consistent?

Upload several images in one message and describe how they combine. Nano Banana 2 treats each upload as a reference: a person, a product, a room, a style. Google says it keeps up to five characters and up to 14 objects recognizable within a single workflow.

This is the most useful feature for anyone making a series. A few ways to put it to work:

• Character in a new scene. Upload a clear photo or drawing of the character and describe the new setting. Refer to the character by what they wear or look like so the model knows which reference is which.
• Product placement. Upload the product and the scene, then say where the product sits and how large it is.
• Style transfer. Upload one image for content and another for style, and say which is which.

Consistency holds best when the reference is clean. A front-facing, well-lit image with a plain background gives the model far less to misread than a cropped screenshot. When a character starts drifting across a long chat, open a new conversation and re-upload the original reference instead of correcting it over and over.

How do you use Nano Banana in Google AI Studio?

AI Studio gives you the same model with more controls, and it is where you go when the Gemini app’s defaults get in the way.

1. Go to aistudio.google.com and sign in with a Google account.
2. Choose Nano Banana 2 (Gemini 3.1 Flash Image) in the model picker.
3. Set the aspect ratio and resolution in the settings panel.
4. Type your description, attach any reference images, and run it.

The API supports 1K, 2K and 4K output for Nano Banana 2, across a range of aspect ratios that includes 16:9 for YouTube frames and 9:16 for Shorts. Images made through AI Studio carry Google’s invisible SynthID watermark, same as the app.

What are Nano Banana’s limits?

The practical limits are resolution, usage caps and age.

• Download size in the app. 1K without a Google AI plan, 2K with one. For 4K, use AI Studio or the API.
• Usage caps. Everyone with a Google account can generate images, but how many depends on your plan, and the allowance refreshes over time. Paid Google AI plans (Plus, Pro and Ultra) get more.
• Editing uploads. 18 and over only.
• Watermarking. Every Nano Banana image carries SynthID, an invisible watermark, and Google attaches C2PA Content Credentials. You can ask Gemini whether an image was made with Google AI.
• Accuracy. It pulls from Gemini’s world knowledge and, for some requests, web search, which helps with real places and infographics. Still check any text, map or diagram it produces before it goes in a video.

Practical tips that fix most bad results

Most failed generations come from the same handful of habits.

Describe a scene, not a tag list. The model understands grammar, so use it.

Put exact text in quotation marks. Nano Banana 2 renders legible text well, and quoting the words tells it exactly what to spell. Keep on-image text short; a four-word title comes out cleaner than a paragraph.

Say the shape first. Cropping a square image to 16:9 later throws away the top and bottom of your composition, which is usually where the headroom was.

Edit in small steps: one change per turn, and restate what must stay the same. And when a long chat starts drifting, stop correcting it. A fresh conversation with the original reference usually fixes a character who has slowly changed faces over twenty turns.

Keep your references. Save the one image that nails your character or style and reuse it every time. That single file is what keeps a series looking like a series.

How do you use Nano Banana for YouTube visuals and thumbnails?

For a longform YouTube video, the fastest way to use Nano Banana is inside a video pipeline rather than one image at a time in Gemini. A ten-minute video can need thirty or more scene images, all in the same style, each matched to the line of narration it sits under. Generating those one by one in a chat, downloading them and lining them up on a timeline is where the time goes.

TubeGen’s AI image generator does that part in one pass. It reads your voiceover, splits it into scenes, generates an image for each one and pins it to the moment it is spoken. Nano Banana 2 is one of the three image models in TubeGen’s Visuals step, next to Grok Imagine and GPT Image, and it is the one TubeGen recommends when every scene has to match one look. It is included on every plan.

How the pieces fit:

• Build a custom art style from your own reference images, or from a YouTube channel link that TubeGen pulls frames from. TubeGen’s model page says Nano Banana 2 follows a saved style more closely than its other two image models.
• Save a recurring host or mascot once with Consistent Characters and it appears with the same identity across every scene and every video.
• You can set Nano Banana 2 for the whole project or only for specific sections. A common setup is a cheaper model as the project default, with Nano Banana 2 switched on for hero scenes where the style has to be exact. It is the highest-credit of the three image models, and the cost is shown before you generate.
• Output is 720p or 1080p inside TubeGen, and it is eligible for batch delivery, where images return within 24 hours.

Thumbnails are a slightly different job. You can make one in Gemini: ask for a 16:9 image with a clear focal subject and space for text, then export it at 1280 x 720, the size YouTube recommends. That works for a one-off. For a channel, TubeGen’s Thumbnail Maker starts from the video’s own title and topic, adds saved characters, objects and text, lets you edit one region without regenerating the whole image, and generates A/B variations before it publishes the thumbnail to YouTube.

The Gemini app is the right tool when you need a single image and want to steer it by conversation. When the job is a whole video’s worth of matching images timed to a voiceover, running Nano Banana 2 inside TubeGen’s model lineup saves the assembly work.

The short version

Open Gemini, click Images, and describe a scene in full sentences. Upload a photo to edit it, one change at a time. Upload several references to blend them or keep a character steady, and move to AI Studio when you need 4K or a fixed aspect ratio.

For YouTube, generate the one-off shots in Gemini and let a pipeline handle the rest. See how TubeGen uses Nano Banana 2 for full videos on its model page.