Google is making the visible watermark on its AI-generated images, videos, and songs optional, the company announced on Friday. The toggle covers outputs from the Nano Banana image model, the Omni video model, and the Lyria audio model, and it will surface first inside Gemini and Google's video editor Flow, with Search support to follow. Invisible SynthID watermarks and C2PA-standard metadata will continue to be embedded in every generation regardless of the setting.
The change addresses a friction point creators have raised since Google started stamping visible marks on generative outputs: the overlay makes the media awkward to use in professional or client work. By keeping the cryptographic layer intact while removing the visible badge, Google is arguing it can serve both audiences at once — designers who need clean deliverables, and platforms that need to verify what is machine-made.
Josh Woodward, Google's VP for Gemini, posted the details on X. Once the feature rolls out in the coming days, users will find the control under Settings > Media Watermark, where the visible mark can be switched on or off per account.
Key facts
- 01Google is making the visible watermark optional on outputs from Nano Banana, Omni, and Lyria across Gemini and the Flow video editor.
- 02Invisible SynthID watermarks and C2PA metadata remain on every generation, so provenance checks in Gemini and Search still work.
- 03Google is open-sourcing Credentio, a library that lets third-party developers embed local provenance validation in their apps.
- 04Support for the toggle in Search is coming after the initial Gemini and Flow rollout, which begins in the coming days.
- 05The change lands days after Anthropic added watermarks to Claude text and file outputs to comply with EU rules.
Provenance tooling stays on the same rails it has been. Gemini and Search can still identify a Nano Banana, Omni, or Lyria file as AI-generated by reading the SynthID payload and the C2PA metadata baked into the asset, whether or not the visible watermark is present. That means an image posted to a social platform or a news site can be checked without relying on a corner logo.
“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. So you can still use Gemini or Search to see if an image was AI-generated.”— Josh Woodward, VP for Gemini, Google
Google is also open-sourcing a new library called Credentio to let developers embed local provenance validation directly in their own apps. That is a meaningful piece: verification today generally requires round-tripping content through a first-party checker like Google's, which limits how quickly third-party platforms — image hosts, social networks, marketplaces — can flag synthetic media at scale. A local library shifts that work to the edge.
The move also sharpens a growing split in how frontier labs are handling AI-generated content disclosure. Anthropic recently added watermarks to text and files produced by Claude to comply with EU regulations, a decision that drew pushback from users who said the marks would out them for using the tool at work or in school. Google is going the other direction on the visible layer for consumer creative outputs, while keeping the machine-readable signals mandatory.
For creators, the practical difference is immediate. A Nano Banana image dropped into a client mockup or a Lyria track cut into a video no longer arrives with a Google-branded stamp attached. For platforms and fact-checkers, the workflow does not change: the SynthID and C2PA layers are what they were already relying on for automated detection, and neither is user-toggleable.
There are limits to how much reassurance the invisible layer can carry on its own. SynthID and C2PA metadata can be stripped or degraded by re-encoding, screenshotting, or aggressive compression, and adoption of C2PA verification outside Google's own surfaces is still uneven. That is part of why Credentio matters — the more apps that can validate provenance locally, the more resilient the system becomes to casual laundering.
The regulatory backdrop is not going away either. The EU AI Act's disclosure requirements for generative content are on the near horizon, and Google's choice to keep the cryptographic layer mandatory while relaxing the visible one looks calibrated to that regime: machine-readable provenance meets the letter of the rule, while creative usability is preserved for paying users.
The bet Google is making is that provenance infrastructure — SynthID at the pixel level, C2PA in the metadata, Credentio on the developer side — is a more durable answer to AI content authenticity than a visible logo the user could crop out anyway. That is probably correct as a technical matter, and it also happens to be the answer that keeps Gemini's creative tools competitive against models that never stamped their outputs to begin with. The next test is whether the platforms that host AI content actually pick up Credentio and start surfacing provenance to end users. Without that consumer-visible layer, invisible watermarks are a policy checkbox more than a public safeguard.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
