Google Adds an Option to Remove Visible AI Media Watermarks
Google will allow users to switch off visible watermarks on AI-generated images, videos and songs, while retaining its invisible SynthID watermark and C2PA-related metadata. Josh Woodward, Google’s vice-president for Gemini, said the option will cover the Nano Banana, Omni and Lyria models.
The control is due to roll out in the coming days through Gemini and Google’s Flow video editor. Support in Search is planned for a later stage. Once the feature is available, users can manage it through Settings > Media Watermark.
Visible labels become optional
The change separates the presentation of an AI-generated asset from the information used to establish its provenance. Visible watermarks can limit the usefulness of media in professional and creative work, while the underlying SynthID signal and C2PA metadata remain intended to support transparency.
Woodward said users will still be able to use Gemini or Search to check whether an image was generated by AI. That means removing the on-image marker does not remove the identification mechanisms Google says it continues to apply.
Models, tools and validation
The initial scope includes Nano Banana, Omni and Lyria, spanning image, video and music generation. It also extends Google’s evolving Gemini product set, including Google’s Gemini Omni video model update, whose Gemini Omni video model is among the products named for the watermark setting.
Alongside the control, Google is open-sourcing a library called Credentio. The library is designed to let developers embed a local validation mechanism in their own applications.
What businesses should do
The announcement arrives as the handling of AI labels remains under scrutiny. Anthropic recently introduced a watermark in text and files generated by Claude as part of its response to EU regulations, illustrating how approaches can differ by media type and provider.
For businesses using Google-generated media, the practical implication is to distinguish visible branding choices from provenance controls: teams can adjust asset presentation while maintaining review processes that use SynthID and C2PA-related signals to identify AI-generated content.