Google Gemini adds DeepMind-powered image-editing tools — and watermarks all AI images
Google has rolled out new image-editing features in the Gemini app powered by DeepMind models. The update—available starting May 2025—adds more advanced, consistent editing tools for both uploaded and AI-generated images.
Key features
- Consistent human appearance across edits: Keep human subjects looking the same as backgrounds, clothing, or settings change.
- Advanced multi-stage editing: Apply sequential edits to different components without losing accumulated changes.
- Image merging and trait-based prompts: Combine two existing images into a new scene, or use a visual trait from one image as a prompt element for another.
- Conversational text prompts: Edit images using natural language within the Gemini app.
Watermarking & ethics
Google noted that all images generated or edited in Gemini include embedded digital watermarks (SynthID) and is experimenting with visible watermarking to increase transparency and help deter misuse.
Background
Image generation of people was temporarily restricted after earlier models produced inaccurate or problematic results for historically specific prompts; users regained the ability to generate images of humans with later models like Imagen 3. The new Gemini tools emphasize visual consistency and iterative control.
Sources & further reading
- Official Google blog: Gemini image-editing announcement
- Tom’s Guide: What the new Gemini editing tools can do
Note: All images edited or generated with Gemini will be watermarked to indicate AI involvement.