Google upgrades Gemini image editing with 'nano banana' model

Gemini is better at editing your images with a new model update from Google DeepMind.

The model, which rose to the top of the LMArena leaderboard under the alias "nano banana," is actually called Gemini 2.5 Flash Image. The Google DeepMind team says this model has been trained to make subjects more consistent across various edits of AI-generated images. This has been an issue for AI image models, given their unpredictable nature.

The ability to upload and natively edit photos in Gemini has been around since April of this year. Instead of learning the technical ins and outs of photo editing software like Photoshop, users can upload an image to Gemini and describe the changes they want in simple terms, no jargon required. With Gemini's updated model, Google says you can do things like change a subject's outfit and location, while keeping their likeness the same.

You can also upload multiple photos and have the subjects appear together in the same photo, or add and alter specific details in an uploaded image to, say, see what a room looks like with a different color of paint or different furniture.

Here's Gemini's attempt at editing my dog into the downward dog pose and relocating her to a yoga studio. Her likeness is the same, and it successfully edited the image to make her eyes open, but her body isn't arched in the way it should be. (I would know, I've seen this playful pose from her many times.)

As Google DeepMind said in its announcement, the model might not always get it right. There might still be inaccuracies with fine details, text in the image, and inconsistencies. With my experiment, my dog's fur looks overly smooth, but her overall coloring, size, and shape stay the same. All images have a visible watermark and an invisible watermark called SynthID to mitigate any confusion over whether they're real or AI-generated. This update is now live, so you can try it out for yourself in the Gemini app.