Edgar Cervantes / Android Authority
TL;DR Users claim the second generation of Nano Banana appeared on a third-party AI media website.
The next version of the image generator is said to use a multi-stage generation process to refine results before outputting a final image.
It’s claimed that Nano Banana 2 will also offer native support for 2K rendering with upscaling to 4K.
Last week, Google released a podcast episode in which it finally explained how its popular image generator, Nano Banana, got its quirky name. While the playful name helped with its rise to fame, it was the tool’s impressive results that made it go viral. Google is reportedly working on a second generation of the AI model, and a leak claims it could offer some significant improvements.
Don’t want to miss the best from Android Authority? Set us as a favorite source in Google Discover to never miss our latest exclusive reports, expert analysis, and much more.
to never miss our latest exclusive reports, expert analysis, and much more. You can also set us as a preferred source in Google Search by clicking the button below.
Users on the site formerly known as Twitter have been claiming that a third-party AI media website “accidentally” provided a preview of what they believe is Nano Banana 2, codename Gempix. These users have shared examples of images produced by the image generator. The examples show improvements in rendering quality, with more detail and fewer instances of AI artifacts.
According to a report from WinFuture, Nano Banana 2 is based on Gemini 2.5 Flash Image. It will reportedly have a couple of standout improvements over the first version of the model. One of those improvements is said to be support for resolutions ranging from 1K to 4K, with native support for 2K rendering and 4K upscaling.
However, a bigger improvement may be the new multi-stage generation process. It’s said that the image generator plans, verifies, and refines results to get rid of inaccuracies and other issues before outputting a final image. This would make it different from most other AI image generators that output an image in one go. The report mentions that this leads to cleaner lines, sharper angles, and fewer errors.
... continue reading