Google Unveils Gemini 2.5 Flash Image: Transform Your Photos with AI Magic!

After previously lagging behind in this field, Google has made a significant leap in image generation. On Tuesday, August 26, 2025, the California-based company officially unveiled Gemini 2.5 Flash Image, a model designed to give users more control over visual editing, “with a particular focus on maintaining character resemblance from one image to another”, addressing a common issue with its competitors. The global rollout began on the Gemini app, Gemini API, Google AI Studio, and Vertex AI, the company stated in a press release.

A model that already dominated LMArena rankings before its release

Developed by Google DeepMind, this model had been the subject of speculation in recent days after being discreetly launched under the codename nano-banana – its internal nickname – on the LMArena platform, which objectively evaluates model performance based on user contributions. Highly effective, it quickly climbed to the top in the Image Edit and Text-to-image categories, notably surpassing 4o Image Generation, a technology integrated into ChatGPT since last March.

strange object spotted under the microscope over the weekend in the lab… pic.twitter.com/t1SBhqAnL0

— Demis Hassabis (@demishassabis) August 25, 2025

Gemini 2.5 Flash Image performs edits without degrading quality

The distinctiveness of this technology, which does not aim to replace Imagen – Google’s other image generation model family – lies in its ability to create new elements or modify existing ones without compromising visual quality or obviously appearing as AI-generated. “We are truly advancing visual quality as well as the model’s ability to follow instructions,” said Nicole Brichtova, product lead at Google DeepMind, as reported by TechCrunch.

With Gemini 2.5 Flash Image, it becomes possible to combine several photos to “create an entirely new scene,” change a person’s outfit or set them in a different background while maintaining their appearance, or “modify specific parts of an image while preserving the rest,” Google illustrates in its blog post. All this can be done with a simple prompt. As demonstrated in the example below, provided by the company, the model can add a sofa, a bookshelf, a rug, or change the paint in an empty room without altering the original photo. The result, it must be said, is quite impressive.

When it comes to editing photos of yourself or people you know, the details matter — a representation that is “close but not quite identical” does not work, says Google.

Google mentions that to prevent misuse, each image created or modified in the Gemini app includes a visible “AI” watermark “as well as an invisible digital watermark SynthID, to clearly indicate they were generated by AI,” they explain.

How to test Gemini 2.5 Flash Image?

Although not available at the time of writing, this “cutting-edge model” has been progressively deployed in the Gemini app since Tuesday, Google states. It will be accessible to both free and paid users, likely with some limitations. Users can simply add one or more images along with a prompt to initiate the generation or editing process.