Google has a new version of its popular image AI. Nano Banana 2.1, released Tuesday, makes pictures from text prompts and edits photos through back-and-forth chat. It replaces Nano Banana 2, which came out in February.

It is rolling out across Google's products, The Decoder reported. Those include the Gemini app, AI Mode in Google Search, Google AI Studio, Flow, Stitch, Google Ads and Google's business AI platform.

The biggest change for builders is price. A standard 1K image now costs developers 3.36 cents, down from 6.70 cents, The Decoder reported. A sharp 4K image falls from 15.10 cents to 7.56 cents. That makes it cheaper for apps to add image tools.

Google says the update improves picture quality, text inside images, and keeping the same characters looking the same across edits. It can combine up to 14 reference images, keeping up to four people and 10 objects consistent, according to The Decoder and Let's Data Science. It can also pull in Google Search results to ground images in real facts, such as for charts and infographics. Users can choose how much the model "thinks" before it draws, which affects quality.

Google also fixed odd tiling marks that showed up on very wide panoramic images, Let's Data Science reported.

Hands-on results were mixed. In The Decoder's own test, Google's pricier Nano Banana Pro still made more natural-looking pictures, even though Google's charts show 2.1 ahead on several scores. In one test image, 2.1 shrank a horse until it looked like a pony.

For developers, there is a deadline. Google will turn off the previous model on Oct. 29, so apps built on it need to switch.

What we know: Nano Banana 2.1 is live across Google products at about half the developer price.
What we don't: How it holds up against rivals in independent testing.