What the Media Watermark toggle does
The change is small in the interface and larger in what it signals about how Google now thinks labeling should work.
Scope of the setting
Where the setting lives, and what it covers
The change was announced by Google's VP for Gemini in a post on X and reported by TechCrunch, not in a post on the company blog, so the detail available is what that post and that reporting contain. The setting covers three models, and it arrives over the coming days, so not every account gets it at the same time.
Check the surface list before you rely on this. Google named Gemini and Flow at launch, with Search support described as coming soon, so output from different Google entry points may not behave the same way during the rollout.
"Josh Woodward, the company's VP for Gemini, said in a post on X that this toggle will be available for Nano Banana, Omni, and Lyria models. He specified that the setting to turn the visible watermark off will be available in Gemini and Google's video editor, Flow, with support for Search coming soon."(the announcement)／"The feature is rolling out in the coming days, and once available, users will be able to go to Settings > Media Watermark to turn the visible mark on or off."(rollout) — TechCrunch
"Gemini Omni Flash is a model that can create anything from any input – starting with video." — Google blog
Why the visible mark was in the way
TechCrunch frames the change as a shift in how labeling is approached: a visible mark often makes the output less useful for professional and creative work, where it is an ingredient in something else rather than a finished post.
That cost is real, and the badge only partly repaid it. A visible mark is also the easiest kind to crop out, which makes it weak evidence in exactly the cases where evidence matters.
"The change reflects an evolving approach to labeling AI-generated media, as visible watermarks can often make content less useful for professional and creative work, but the need to identify AI-generated content remains." — TechCrunch
What stays in the file
Two provenance mechanisms survive the toggle, and they work in different ways. Neither is visible, and neither is under the user's control.
Three layers, one switch
SynthID is embedded and stays
Google adds SynthID at the moment content is created. It sits inside the media itself, and Google states it does not change image or video quality. It is imperceptible to people and readable by Google's detection tooling.
SynthID is why Google can afford to make the badge optional. Removing a mark from the surface of an image does not touch a signal carried in the pixels, so the detection path Google points users toward keeps working either way.
"SynthID embeds digital watermarks directly into AI-generated images, audio, text or video. The watermarks are embedded across Google's generative AI consumer products, and are imperceptible to humans – but can be detected by SynthID's technology."(How SynthID works)／"SynthID adds an invisible digital watermark to an AI-generated image (or video segment). The watermark doesn't change the image or video quality. It's added the moment content is created"(How SynthID works) — Google DeepMind
C2PA metadata travels with the file
C2PA is the other half. It is an open standard for recording where a file came from and how it was edited, carried alongside the file rather than inside the pixels.
Google's stated position is that both signals stay on whatever the toggle is set to, and that Gemini and Search can still tell you whether an image was AI-generated.
"The Coalition for Content Provenance and Authenticity, or C2PA, provides an open technical standard for publishers, creators and consumers to establish the origin and edits of digital content."(what C2PA is)／"Content Credentials function like a nutrition label for digital content, giving a peek at the content's history available for anyone to access, at any time."(Content Credentials) — C2PA
""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," he wrote." — TechCrunch
Moving verification outside Google
What will outlast the toggle is a change in who gets to do the checking.
Where checking happens
Credentio and local validation
Google is open sourcing a library that puts a validation mechanism inside third-party applications. That moves the check from a service you query to a component you ship.
If you build anything that ingests user media, Credentio matters more than the toggle. A local check runs on upload, at whatever volume you handle, and does not depend on an external endpoint staying up.
"Google is also open sourcing a new library called Credentio to let developers embed a local validation mechanism in their apps." — TechCrunch
What to assume about a file with no visible mark
The practical rule changes in one direction only. An image without a badge is no longer evidence that it was not generated by Gemini, and it never was strong evidence in the first place.
Going the other way still works. On the statements quoted above, a SynthID hit or intact C2PA data remains a positive signal, so provenance checks should read the file rather than look at it.
For the same question at a different lab, see Claude Text Watermark: What a Hit Actually Proves. For the transparency rules driving much of this work, see EU AI Act: transparency enforcement.
Vendor announcements and standards documentation are easier to compare once both are plain text.
