How to Use Nano Banana: Google’s Image Model, Explained Step by Step

The most confusing thing about learning how to use Nano Banana is that the words “Nano Banana” appear nowhere in the app you are supposed to use it in. Google’s image model has a nickname that went viral, an official name almost nobody says out loud, and a Gemini interface that labels it by neither. So people open Gemini, hunt for a banana, find nothing, and assume they need a different app.

You do not. Nano Banana is the engine behind image generation in Gemini, and reaching it takes about twenty seconds once you know which menu it hides in. This guide covers where it lives on web and mobile, and which of the four Nano Banana versions you are actually running. It also covers how to edit your own photos with it, and what to do when a result comes back wrong.

The Key Takeaways

• It is never called Nano Banana in the app: look for Create image in the plus menu, or Images in the web sidebar.
• Four versions exist: the original, Nano Banana 2, Nano Banana Pro and Nano Banana 2 Lite. Which one you get depends on your chat model setting, not on a separate image picker.
• Nano Banana 2 is the default and is officially Gemini 3.1 Flash Image, released in February 2026. It handles 512px to 4K and holds up to five characters and 14 objects consistent.
• Editing is the strongest use case: upload a photo, describe the change in plain language, and change one thing at a time.
• Free accounts get 1K downloads, a Google AI plan raises that to 2K and unlocks regenerating with Pro.

How to Use Nano Banana in Under a Minute

Here is the short version. Everything after this section explains the parts that trip people up.

1. Open the Gemini app on your phone, or go to gemini.google.com in a browser, and sign in with any Google account.
2. Tap the plus (+) button next to the prompt box and choose Create image. On the web you can also open the sidebar and pick Images.
3. Type what you want to see. To edit a photo instead, upload it first, then describe the change.
4. Wait a few seconds for the result, then reply in the same conversation to refine it.
5. Use the three dot menu to download, or to regenerate the image at higher quality if your account includes that.

That is the whole loop.

No settings to change, no model to install, no separate app to download. The friction is entirely in step two, because the menu has moved twice in the past year and most guides online still describe the old layout.

Where Nano Banana Actually Lives

Google has never shipped a button that says Nano Banana. The name started as an internal codename attached to an anonymous model on the LMArena leaderboard in August 2025. It went viral, Sundar Pichai confirmed it with three banana emojis, and it stuck. Officially the models are Gemini image models. In the interface they are just image generation.

Worse, the entry point moved. Until March 2026 there was a separate Tools menu holding image generation. Google folded that into the plus menu on the web, so guides written before then send you to a menu that no longer exists. If you have been following instructions that mention a Tools menu and getting nowhere, that is why.

On the web

Two routes work. Click the plus (+) icon beside the prompt box and select Create image, which keeps you in the current conversation. Or open the sidebar and choose Images, which drops you into a dedicated view with prompt templates. The sidebar route is better for starting fresh, the plus menu is better when you are already mid conversation and want an image based on what you have been discussing.

On iPhone and Android

Tap the plus icon to the left of the prompt box, then Create image. Uploading a photo to edit works the same way: attach it from your camera roll, then type the change you want in the same message. On Android you can also hold the power button to launch Gemini and go straight to a prompt, which saves a few taps when you want to edit a photo you just took.

Which Nano Banana Are You Using?

This is the part that generates the most confusion, and the part almost no guide answers, because the answer changed twice this year. There is no image model picker. The image model you get is decided by whichever chat model is selected at the top of your conversation.

As of August 2026 that picker reads 3.7 Flash, 3.5 Flash-Lite and 3.1 Pro. Those are chat model names, and each one routes to a different image model behind the scenes. According to 9to5Google, 3.7 Flash replaced 3.6 Flash across Android, iOS, web and macOS in mid August 2026, so if you read a guide referring to Fast and Thinking options, it predates the rename. Our breakdown of what shipped in Gemini 3.7 Flash covers that release on its own terms.

Comparison At Glance

In practice most people are on Nano Banana 2 without knowing it. Google’s own announcement post puts its output range at 512px to 4K, with accurate in-image text and consistency across up to five characters and 14 objects in a single scene. That last figure matters more than it sounds: it is what lets you put the same person in six different settings and have them still look like the same person.

Pro is the slower, more deliberate option, built on Gemini 3 Pro rather than a Flash model. Reach for it when a scene has complicated spatial relationships or a lot of text that has to render correctly. Our writeup of what changed in Nano Banana Pro goes deeper on where the extra compute shows up, and the Nano Banana 2 launch breakdown covers the February release in full.

How to Edit Photos With Nano Banana

Generation gets the attention. Editing is where the model earns its reputation. You upload a photo, describe a change in ordinary language, and it rewrites only the part you asked about while leaving faces, lighting and background intact.

The rule that makes editing work: change one thing per message. Stacking three edits into one prompt is how you get a result where the background is right and the face has quietly become someone else.

Start with a background swap, because it is the easiest to judge. Upload the photo and write “Replace the background with a plain white studio backdrop and soft shadows. Keep the subject, pose and lighting exactly as they are.” Object removal follows the same shape: “Erase the bin on the left and fill the space to match the surrounding pavement.” Things with clear edges disappear cleanly, while anything overlapping your subject takes a few attempts.

Colour and garment changes are where the model shows off. “Change the jacket to dark green. Do not change the face, hair or background.” Restoration is the other party trick, and it needs the same discipline: “Repair the scratches, sharpen the detail and colourise this photograph naturally. Keep the original composition.”

Notice the pattern in all four. Say what changes, then say explicitly what must not, because that second half is doing most of the work here. If you want a bigger library to work from, our roundup of ten Nano Banana examples with copy-paste prompts is built for exactly this.

Where Else Nano Banana Runs

The Gemini app is the main door, not the only one. The same models power image generation across a spread of Google products, and a few of them suit particular jobs better than the chat window does.

AI Mode in Search and Google Lens handle quick one-off images without making you leave a search. Google AI Studio is the better choice for higher volume work and finer control, since it behaves more like a developer console than a chat app. NotebookLM builds illustrations into slide decks and infographics generated from your own sources, and Google Slides and Vids reach the same models through the Gemini sidebar so visuals land straight in a deck. Flow and Google Ads cover video and campaign creative on paid plans.

One quota note before you plan around any of this. Google’s official help documentation states that once you hit your daily Nano Banana 2 limit, you lose the ability to redo images with Pro as well. The two allowances are linked, so burning through fast generations early costs you the good ones later. For the actual daily numbers and what the free tier includes, see our breakdown of whether Nano Banana is free and where the limits sit.

How to Use Nano Banana on a Mac

There is no standalone Nano Banana app for macOS, and there will not be one. The models run in Google’s cloud and are reached through whatever client you use to talk to Gemini. On a Mac that leaves three options.

The browser works fine and needs no setup. Google’s native Gemini app for Mac, released in April 2026, adds an Option and Space shortcut that pulls Gemini over any window, which saves real time on quick edits. The catch is the requirements: Apple Silicon only, macOS 15 Sequoia or later, so Intel Macs are out. We covered the details in our Gemini Mac app review.

The third route is a multi-model Mac client like Fello AI, which runs on Intel Macs as well as Apple Silicon and works on iPhone and iPad from the same account. It keeps Gemini alongside other image and chat models in one window, so you are not switching apps when you want to compare a Nano Banana result against another model. If you are weighing up plans first, our guide to Gemini pricing lays out what each tier actually includes.

When Nano Banana Gets It Wrong

Four failures account for most bad results, and all four have fixes.

If the face changed, you asked for too much at once. Undo, then make the single edit you care about with an explicit instruction to leave the subject untouched. If the text came back as gibberish, blame the model tier. Flash is built for speed, not precision. Regenerate with Pro when your plan allows it, since accurate lettering is the clearest thing the larger model buys you.

If your download looks soft, check your plan. Free accounts download at 1K, a Google AI plan lifts that to 2K, and generation itself reaches 4K on the right model.

If nothing happens at all, the cause is usually eligibility rather than a bug. On personal accounts, generating and editing images is restricted to users 18 and over, while generation alone opens up at 13. Workspace accounts follow different rules. Availability also varies by country and language, so a feature that works for someone else might not have reached your account yet.

One more thing to know before you publish anything you make. Google added a setting in August 2026 that turns off the sparkle watermark in the bottom right corner of generated media. You will find it under Settings, then Media Watermark, then Show watermark. It does not make the image untraceable. Invisible SynthID marking and C2PA metadata stay embedded either way, and the toggle is unavailable in countries where visible labelling is a legal requirement.

The Bottom Line

Nano Banana is easier to use than it is to find. Once you know it lives behind Create image in the plus menu, most of the difficulty disappears. Your chat model quietly decides which version you get. After that it is a short loop: describe, generate, refine one detail at a time.

Start with editing rather than generation. Upload a photo you already care about, change the background, and you will learn more about how the model reads instructions in two minutes than any prompt library will teach you. Then pick your surface: the browser for occasional use, a native Mac client if you are reaching for it daily.

FAQ