Google is loosening one of the most obvious labels on AI-made media: Gemini users will soon be able to switch off visible watermarks on generated images, videos and songs. The company insists the change does not mean those works will become untraceable.

Google announced Friday that the option will cover its Nano Banana, Omni and Lyria models, reflecting a shift from conspicuous labeling toward behind-the-scenes verification. The toggle is due to appear in Gemini and Google’s Flow video editor in the coming days, with Search support planned next; users will find it under Settings > Media Watermark.¹

The practical appeal is clear. Visible marks can limit the usefulness of AI outputs in professional and creative settings, while creators who want a cleaner final product may see the label as an unnecessary constraint. But the policy also lands amid wider concern that removing obvious signals makes synthetic media harder for ordinary viewers to spot.

Google’s answer is to retain invisible SynthID watermarks and C2PA-related metadata, which can be checked through Gemini or Search. Josh Woodward, Google’s vice president for Gemini, framed the move as a compromise: “We’re striking a balance here between creative control and safety: while the visible watermarks are now optional, invisible SynthID watermarks and C2PA metadata are still being used for transparency.”¹

The company is also open-sourcing Credentio, a library intended to let developers build local validation tools into their own apps. The move follows Anthropic’s contested decision to watermark Claude-generated text and files in response to EU requirements—evidence that the fight over AI labels is moving from whether they should exist to how visible they must be.¹