Google on Tuesday, Oct. 6, released Nano Banana 2.1, an update to the AI image maker that turns a written request into a picture or edits a photo you give it. Google says the new version beats its earlier image models across the board, with notable gains in design, in editing parts of a picture a user marks, and in keeping the same people and objects looking the same from one image to the next. For businesses that make pictures in bulk, the price may matter more: each finished image now costs developers about half what it did with Nano Banana 2.
Google lists the new model in the Gemini app, AI Mode in Google Search, Google Ads, Google AI Studio and two other Google creative tools, Flow and Stitch. Developers can also build it into their own apps, and Netlify, a web hosting company, already offers it to its customers.
On Google's paid plan for developers, a standard picture of about 1,000 by 1,000 pixels costs about 3.4 cents, down from about 6.7 cents with Nano Banana 2, and the largest 4K pictures cost about 7.6 cents, down from about 15 cents. There is a catch. The written instructions and photos developers send in now cost three times as much as before, and the text the model writes back, including the planning it does before drawing, costs two and a half times as much. Jobs that send long instructions or many photos, or that ask the model to plan longer, will save less. As with Nano Banana 2, there is no free developer plan.
Developers can give it up to 14 reference images, photos for the model to work from, and Google says it can keep up to four characters and up to 10 objects looking the same in the pictures it makes from them. That would help a business that wants the same product or person to appear across a series of pictures, such as an ad campaign or a storyboard. Google also says the model writes words inside pictures more clearly, lays out infographics more accurately, and fixes flaws that showed up in very wide or very tall pictures at the larger sizes. Like the version before it, it can look things up in Google Search and Google Images before it draws. Developers can also set how much planning the model does before it draws, from minimal to high.
Google's claims about quality rest on its own tests. In them, people comparing pictures made from written requests side by side preferred Nano Banana 2.1 over both Nano Banana 2 and Nano Banana Pro, Google's pricier image model, and it rated highest when it planned before drawing. Google also lists the model's limits: small text is often blurry at the standard size, characters do not always match the photos they come from, and it sometimes mixes up left and right.
Google has not said whether everyone using the Gemini app gets the new version, including people on the free plan. It is also not yet clear what will happen to Nano Banana 2 for developers who built their products on it.