Google DeepMind has introduced a significant update to its Gemini AI, rolling out a suite of new image editing features. These tools, available now in the Gemini app, are designed to offer users more powerful and consistent ways to create and modify images. In line with industry-wide efforts for transparency, Google confirmed that all images either generated or edited by Gemini will carry a visible and digital watermark to clearly identify them as AI-made.
A core focus of the new model is maintaining consistency, especially when editing human subjects. The AI is now better at keeping a person’s appearance the same across multiple edits to the surrounding scene. This allows a user to upload a photo of someone and then use AI to realistically place that person into different backgrounds or change their outfit, all while ensuring the individual remains recognizable and true to their original likeness.
Beyond consistency, the update brings a set of advanced editing capabilities. Users can now merge two separate, pre-existing images to craft an entirely new scene. Another feature allows a visual element from an existing picture to be used as a new prompt or a design component for generating fresh content. The model also supports multi-stage editing, which lets users make a sequence of changes to individual parts of an image without losing any of the previous adjustments.
This expansion follows a period where Gemini had disabled its ability to generate images of people. The feature was paused last year after the model produced historically inaccurate and overly diverse representations in response to prompts about specific historical periods and figures. The capability to generate human images was later restored with the launch of the upgraded Imagen 3 model.
