Google Gemini: New AI Model for Images

Image editing functionality has been available in the Gemini app since the end of April . Google announced a major update today. All users (even those without a subscription ) can use the new Gemini 2.5 Flash Image model to edit images.

World's best text-to-image template

Image editing in the Gemini app was introduced on April 30. Previously, it was only accessible to developers via AI Studio and the Gemini API. It leveraged the Gemini 2.0 Flash model. The new Gemini 2.5 Flash Image model, currently ranked #1 in LMArea , offers significant improvements to both image generation and editing.

Image generation with Gemini just got a bananas upgrade and is the new state-of-the-art image generation and editing model. 🤯

From photorealistic masterpieces to mind-bending fantasy worlds, you can now natively produce, edit and refine visuals with new levels of reasoning,… pic.twitter.com/hYwA6l4QyY

— Google DeepMind (@GoogleDeepMind) August 26, 2025

To use the new model, simply upload an image and unleash your imagination and creativity, entering a text prompt to indicate the desired final result. The primary goal of Google DeepMind experts is to preserve the subject (person or animal) in the original image.

Users can perform traditional edits, such as changing the background or adding an object, by typing a detailed text prompt. Google has published some more complex examples. You can ask Gemini to change the subject's clothing or location in the photo:

You can also ask to generate a new image by combining two photos. This is the final result:

The new model also allows batch editing on the same photo uploaded to Gemini. For example, a user can start with a photo of an empty room and ask it to sequentially change the wall color and add a bookcase, a sofa, a rug, and a coffee table.

The last example shows applying the style of one image to the object in another image:

All images created or modified will be watermarked with SynthID technology. Google has implemented a filter that prevents the generation of non-consensual intimate images.