The Mysterious Nano-Banana Image Model is From Google and It’s Rolling Out in Gemini

For weeks, the mysterious nano-banana image model has been earning praise for generating impeccably accurate images. It could realistically blend images, edit parts of images while maintaining incredible coherence, and transform photos with simple natural language prompts. As it turns out, the nano-banana image model is actually Google’s Gemini 2.5 Flash Image AI model.

Yes, you read that right. Google was the first company to showcase Gemini’s native image generation in March and it was powered by the Gemini 2.0 Flash Image model. Now, the upgraded native image generation feature is powered by Gemini 2.5 Flash Image. Unlike Diffusion models, it’s natively multimodal and maintains incredible accuracy over multiple generations.

The new Gemini 2.5 Flash Image (nano-banana) is currently the top-rated model on LMArena for image editing. It has scored a whopping 1,362 ELO points, much higher than the second best, Flux.1 Kontext [max] which has received 1,191 points.

It goes on to show that Google’s Gemini 2.5 Flash Image model is top-notch and can edit images conversationally while preserving the scene. What is interesting is that Google is already integrating the new image model in Gemini. Starting today, you can upload and edit images in Gemini and it will use the new Gemini 2.5 Flash Image (nano-banana) model.

You can use this feature to combine multiple images, change specific parts of the image using text prompts, reimagine yourself in different avatars, and more. You can even try new costume, change the location, mix up images, and perform multi-turn editing over and over again while preserving the overall scene.

Note that all images generated in the Gemini app will have a visible ‘ai’ watermark and an invisible SynthID watermark.