A mysterious AI-powered image generator, Nano-Banana, went viral, sparking speculation that it could have been created by Google. Many testers thought it was released on LMArena as a test version before its full rollout, as is often done with next-generation AI models.
Today, Google officially announced that Nano-Banana is its next-gen AI image generation model. Its real name is Gemini 2.5 Flash Image, but Google acknowledged that most people will use the nickname.
According to Google, both paid and free Gemini subscribers can use Nano-Banana in-app image editing via the Gemini app. The upgraded version will also be available in the Gemini API, Google AI Studio, and Vertex AI.
Google revealed in a blog post that the new model comes from its DeepMind department. The company has said that people have already been going bananas on it in early previews, and this model is currently the top-rated image editing system globally on LMArena ratings as of today.
Gemini 2.5 Flash Image builds on Google’s previous AI-image generation features released this year. The main improvement is scene and character consistency.
Google explained
“Our latest update is designed to ensure that photos of your friends and family, and even your pets, look consistently like themselves, as you try to place a 60’s beehive haircut or putting a tutu on your chihuahua.”
Users can give Gemini an image with a text prompt, and the AI will make edits to the image in the likeness. It can also merge two photos or convert parts of one photo into another.
Another standout feature is the multi-turn editing, whereby a user can edit an image repeatedly without losing consistency. For example, starting with a plain room, the AI can recolor the walls, then add a bookshelf, a couch, and a rug, all while maintaining the same perspective and scene details.
