Google Makes Visible AI Watermarks Optional, Keeps Invisible Ones

Google is updating its AI tools to let users turn off visible watermarks on content generated with Gemini and its Flow video editor. The change, announced Friday, applies to the "sparkle" watermark that appears in the bottom-right corner of images, videos, and music created with the company's Nano Banana, Omni, and Lyria models. Users can toggle the new "Media watermark" setting in Gemini and Flow, with support for Search coming soon.

The move is part of a broader shift in how AI-generated media is labeled. As Josh Woodward, VP of Google Labs, Gemini, and AI Studio, explained, the goal is to balance creative control with safety. While visible watermarks are now optional, Google will still embed invisible SynthID watermarks and C2PA metadata in the background. This means AI-generated content remains verifiable via tools like Gemini or Search, even if the visible cue is removed.

The feature is rolling out in the coming days, but Google will not offer the toggle in countries that require visible watermarks. Also, the company is open-sourcing a new library called Credentio, which lets developers embed a local validation mechanism in their apps.

This update follows Anthropic's controversial move to add watermarks to text and files generated by Claude, citing EU regulations. Google's approach contrasts with OpenAI and Meta, which already avoid visible watermarks on AI-generated images, relying instead on invisible markers like SynthID or their own standards.

For developers and content creators, this change means cleaner outputs for professional use, but it also raises questions about how easy it is for the average user to tell if something is AI-made. The invisible markers provide a technical solution, but the practical impact on misinformation remains to be seen.