Google just upgraded its AI image model, and it actually looks to be a pretty significant step up. The company rolled out Gemini 2.5 Flash Image today, a major refresh that promises smarter and more flexible image generation.
The upgraded model allows users to issue natural language prompts to not only generate images but also merge existing photos and make more precise edits without creating weird distortions. It also taps into Gemini’s “world knowledge” to better understand what it’s generating.
This upgrade comes as Google tries to close the gap with the industry leader, OpenAI. In the past, image generation has been a major driver for AI. ChatGPT usage skyrocketed in March when the company launched its GPT-4o native image generator. The viral Studio Ghibli memes generated by the model resulted in the company’s GPU models melting, according to OpenAI CEO Sam Altman. ChatGPT currently has over 700 million weekly users. By comparison, Google CEO Sundar Pichai revealed on the company’s July earnings call that Gemini had 450 million monthly users, still trailing behind ChatGPT.
With its latest update, Google says it’s solved one of AI’s biggest headaches. Until now, keeping characters or objects consistent across multiple edits has been a major challenge for AI image generators.
“You can now place the same character into different environments, showcase a single product from multiple angles in new settings, or generate consistent brand assets, all while preserving the subject,” the company wrote in a blog post.
Google says users can now make very specific tweaks with just a prompt. For example, users can blur the background of an image, remove a stain from a T-shirt, change a subject’s pose, or even add color to a black-and-white photo.
Even before its official launch, the new model was turning heads on the crowdsourced evaluation platform LMArena, where it appeared anonymously under the name “nano-banana.” One X user shared how they used nano-banana to change Altman’s shirt in a photo. The result was surprisingly good. Today, Google stepped forward and claimed ownership of the model, revealing that nano-banana was in fact Gemini 2.5 Flash Image.
In addition to being available on the Gemini app, the new model is now accessible to developers through the Gemini API, Google AI Studio, and Vertex AI. Google has already built several template apps that make use of the new model on Google AI Studio, the company’s coding AI assistant, and said users can vibe code on top of them.
The company also said some developers have already experimented with the app to see how it would be useful in real-world scenarios, like creating real estate listing cards, employee uniform badges, and product mockups.