AIBRIEF
The setting rolls out in Gemini and Flow, Google's video editor, in the coming days, while Google keeps SynthID (its invisible watermark) and C2PA content credentials metadata on every file, and open sources Credentio, a C++ library that lets
Google will let users turn off the visible 'AI-made' label on its generated images, video, and music. The setting rolls out in Gemini and Flow in the coming days, with Search support to follow, and lives at Settings > Media Watermark.
The invisible layer is not affected. SynthID stays on every generation. The C2PA content-credentials metadata, a provenance standard that lets a file carry its own origin record, stays on too. Google says SynthID has now been used to watermark more than 10 billion pieces of content.
VP of Gemini Josh Woodward named the in-scope products on X: the Nano Banana and Omni image models, and Lyria for music.
Google is also open-sourcing Credentio, a C++ library for C2PA content credentials, so third-party apps can validate the invisible layer locally without round-tripping through Google.
The move positions Google against a sector going the other way: Anthropic embeds watermarks in Claude text and files to comply with EU rules, while Google is making the visible label the part users can opt out of and expanding the invisible layer's portability.
What remains unknown: how many users will toggle the visible label off, and whether third-party apps will adopt Credentio fast enough to make the invisible layer a real consumer check.