Google lets Gemini users turn off the visible watermark on AI-generated content

Google has begun letting users switch off the visible watermark stamped on content made with its Gemini AI app, while keeping an invisible watermark and provenance data embedded underneath.

The new “Media Watermark” setting, in the Gemini app and its Flow video editor, turns off the visible mark on AI-generated images from Nano Banana, video from Omni and music from Lyria; support in Google Search is coming. Gemini vice president Josh Woodward said the change was rolling out.

Google’s position is that transparency survives the toggle. Its invisible SynthID watermark and C2PA provenance metadata stay embedded in generated files whether or not the visible stamp is shown, so the origin of a file can still be checked by tools that read those signals. Google is also open-sourcing Credentio, a C2PA validation library for developers.

The catch is that invisible provenance only helps people who have a detector, while the visible watermark is the one an ordinary viewer actually sees. Removing it makes AI images, video and music harder to spot at a glance, at a moment when synthetic media is spreading fast.

Access varies. In India, South Korea and Vietnam, switching the watermark off requires a paid AI Ultra subscription, and work and school Google accounts do not get the setting at all. That leaves the strongest visible labeling in place for enterprise and education users while consumers can opt out.

Whether regulators reading the EU AI Act’s labeling rules accept invisible SynthID as sufficient, or push back on hiding the visible mark, will shape how far this design travels.