
Google's Nano Banana 2.1 Arrives in Gemini and Search
Google released Nano Banana 2.1, the new version of the AI that makes and edits pictures in the Gemini app, and it's rolling out in Gemini and in Google Search's AI Mode.
Google released Nano Banana 2.1 on Tuesday. Nano Banana is Google's name for the AI that makes and edits pictures in the Gemini app, and 2.1 is the new version. Google says it's better than its earlier image models across the board, with the biggest gains in design, editing one part of a picture, and keeping people looking the same from one image to the next.
It's rolling out in the Gemini app, AI Mode in Google Search, Google's video tool Flow, its design tool Stitch, and Google's ad tools.
What's new
2.1 replaces Nano Banana 2, the version Gemini has used since February. Google's developer docs call it the more efficient counterpart to Nano Banana Pro, a separate, higher-end image model that paid Gemini users can switch to.
Three changes matter most if you make images for work:
- The same people across a series. You can give it up to 14 reference images, and it keeps up to four people looking the same across the pictures it makes, plus up to ten objects, like a product or a mascot. That's what you need for a set of social posts or a storyboard with recurring characters. Those limits come from Google's developer docs; Google hasn't said how many photos the Gemini app itself takes at once.
- Change one part, keep the rest. You can describe in plain words the one thing to change, like a shirt or the sky, and it leaves everything else alone. Google's image-editing guide suggests phrasing it as "change only the [specific element] to [new element/description]. Keep everything else in the image exactly the same."
- Better text and layouts. Google says it renders text more accurately and lays out infographics, posters and diagrams better than before.
It also makes images at up to 4K and fixes glitches that showed up in very wide or very tall pictures. Google is also shutting off the old Nano Banana 2 for developers on October 29 and telling them to switch to 2.1.
How to try it
The quickest way is in Search. Open the Google app on your phone and tap the banana icon under the search box. Robby Stein, who runs product for Google Search, posted the how-to when it went live in AI Mode.
AI Mode is part of regular Google Search, and Google's post about it mentions no subscription.
In the Gemini app, it's part of a gradual rollout, and not everyone has it yet. Google hasn't said which Gemini plans get 2.1 or whether free users do, and its Gemini help pages still describe Nano Banana 2.
How it stacks up against Nano Banana Pro
Google says 2.1 beats it. On the model card, 2.1 scores ahead of both Nano Banana 2 and Nano Banana Pro in Google's own tests of people rating images side by side.
One early hands-on test disagreed: in a side-by-side, Pro's picture of a horse looked more realistic and natural, and 2.1's came out closer to a pony.
Some early users say the realism is a step up, though text on signs can still give it away.
Google's own list of limits on the model card includes the text problem. It says small text can come out blurry, characters don't always match their reference photos, marks you draw to guide an edit sometimes stay in the picture, and it occasionally mixes up left and right.
How it compares with ChatGPT Images 2.5
I haven't tried 2.1 yet. ChatGPT Images 2.5 makes all my thumbnails and most of my photos, and it's been exceptional. It's also the model behind ChatGPT's new Try On button for clothes. So the question for me is how much better Nano Banana really is.
Going only on Google's published details, here's where 2.1 could pull ahead:
- Recurring characters. Four consistent people and 14 reference images is built for a series, like a week of posts with the same cast.
- Small fixes to an image you like. Changing one element while leaving the rest untouched is what you want when an image is nearly right.
- Where it lives. It's built into Google Search and the Google app, which you may already have open.
On the ChatGPT side, OpenAI is about to test picture ads while images generate for Free and Go users in the US.
And if most of your pictures end up on Instagram or WhatsApp, Meta has its own free image model, Muse Image, built into both.
I put the earlier version, ChatGPT Images 2.0, through more than 50 use cases:
One early side-by-side test found Pro's picture more realistic, and a lot of early users say 2.1 follows instructions less closely.
The banana icon in the Google app makes it easy to compare with ChatGPT using the same prompt and the same small fix.
Sources
- @Google on X: Meet Nano Banana 2.1
- @Google on X: where it's rolling out
- Robby Stein on X: Nano Banana 2.1 in AI Mode
- Google DeepMind: Nano Banana 2.1 model card
- Google AI for Developers: Nano Banana 2.1
- Google AI for Developers: image generation guide
- The Decoder: Nano Banana 2.1 hands-on
- Android Authority: Google launches Nano Banana 2.1



Comments
Sign in to join the conversation.