A mysterious Nano-Banana AI image generator went viral a few days ago, with some testers speculating that Google owned the new generative AI tool. They thought Google released it on LMArena for public testing before the official launch, which is what usually happens with new frontier AI models that developers want to offer for testing before a wider rollout.
Google on Tuesday confirmed that Nano-Banana is indeed its next-generation AI image generation model for Gemini. Its official name is Gemini 2.5 Flash Image, though you won’t have to bother remembering that name. According to Google, paid and unpaid Gemini subscribers can use the “Nano-Banana” image editing capabilities in the Gemini app. The upgraded AI image generator will also be available in the Gemini API, Google AI Studio, and Vertex AI.
Google
The new editing model comes out of Google’s DeepMind department, Google said in a blog post. “People have been going bananas over it already in early previews,” Google teased, adding that the new model is the top-rated image editing model in the world, according to LMArena rankings.
It’s all about character consistency
Gemini 2.5 Flash Image is the continuation of Google’s existing AI image generation capabilities released in the Gemini app earlier this year. The focus of this update is on character and scene consistency, according to Google. “Our latest update is designed to make photos of your friends, family, and even your pets look consistently like themselves, whether you’re trying out a 60’s beehive haircut or putting a tutu on your chihuahua,” Google said.
Google
All you have to do is feed Gemini an image and a prompt, like the one above, and wait for the AI to process it. The new Gemini 2.5 Flash Image model will keep your likeness in place and make the edits you wanted. You can also instruct the AI to combine two different images into a single photo or adapt elements from one image to create another. The following example shows a butterfly dress imagined by Gemini AI after it was told to use a butterfly as an example.
Google
The new AI image generator can also support multi-turn editing and maintain the consistency of a scene. Google offers an example of using the photo of an empty room to imagine it in a different paint color. The user then asks the AI to place a bookshelf, a couch, and a rug in the room, and Gemini does it without changing the overall image.
Safety features in Gemini AI images
Google’s examples also show that you can tell the AI to do whatever you want to an image. You can change the background or edit a single detail in the photo. You can place yourself in any photo you can imagine, in any style you want. You can grab the design style of an image, like the texture and color of flower petals, and apply it to a pair of boots. That’s actually an example that Google uses to showcase Gemini’s new image generation and editing skills.
The AI will create images that look like they’ve been captured with a real camera. The Nano-Banana samples that went viral a few days ago already proved that point. It’s no surprise that Gemini 2.5 Flash Image now tops LMArena benchmarks.
Google
However, such capabilities also open the door to abuse. Some people might want to use these powerful AI image editing tools for malicious activities, like misleading others. The good news is that the Gemini app will include a visible “AI” watermark on all images created and edited in the Gemini app. All photos will also feature the usual, invisible SynthID digital watermark to indicate they were generated with AI.