Google has just introduced its new image editing model developed by its DeepMind division, integrating it into the Gemini app. It will give users an option to modify and create pictures in more advanced ways just by a single prompt. Taking to the blog post, Google stated that the update will be available starting today, and it aims to improve how consistently the app represents people, pets, and familiar subjects during edits, addressing a longstanding challenge with AI-generated images.
This comes after Gemini earlier got the native image editing features. The blog post further mentioned that the new model aims to maintain the likeness when making changes to photos, a problem which often results in inaccuracies, particularly when the users try to make changes to the hair, outfits, expressions, and more. Google says the update should reduce such inconsistencies, ensuring that edited images still resemble the original subject. To use it, the users need to give a prompt and image to Gemini.
Additionally, the model also supports new features. The users will be able to upload multiple photos and blend them into a single image, such as combining your portrait with a pet’s picture to make it a shared scene. The model also supports multi-step or “multi-turn” editing, allowing users to iteratively modify different elements within the same image. For instance, you can repaint the walls of a digital room, then add furniture, paintings, and much more, without losing earlier adjustments.
Also read: Apple iPhone 16 Plus price drops to under Rs 65,000 ahead of iPhone 17 series launch
The company also stated that the model will let users transfer visual styles across objects, such as applying the texture of flower petals to clothing or borrowing patterns from one image to redesign another. Background changes and costume alterations are also possible, while retaining the subject’s original appearance.
The tech giant has already confirmed that all the images that will be produced or edited via the Gemini app will get a visible watermark, an invisible SynthID digital watermark to identify them as AI-generated.