Google improves Gemini AI image editing with “nano banana” model (arstechnica.com)

Google has unveiled its latest advancement in AI image editing by integrating the "nano banana" model—officially known as Gemini 2.5 Flash Image—into the Gemini app. Developed by Google DeepMind, this model quickly rose to the top of the LMArena leaderboard for image editing, thanks to its breakthrough ability to maintain consistency and fine details across multiple edits. Unlike prior generative models prone to unpredictable changes, nano banana “remembers” key elements of the source image, delivering reliable and coherent transformations even after several modifications. This consistency is a significant leap for AI-driven image editing, broadening practical use cases that demand visual continuity. For instance, users can upload a photo of a person and apply diverse stylistic changes—such as altering clothing or themes—while preserving the individual's core features throughout each edit, whether reimagining them as a matador or a '90s sitcom character. This innovation not only enhances creative control but also reduces the trial-and-error frustrations common in generative AI workflows, making it an important milestone for artists, designers, and developers leveraging AI to streamline and expand visual content creation.