Google is finally letting you turn off the visible watermark on Gemini-generated images, videos, and music. The new Media watermark setting is rolling out in the coming days, and it puts the sparkle badge in your control instead of Google’s.
What the Media watermark toggle does
The Media watermark setting lives in Gemini’s settings menu. Turn it off and future generations skip the visible sparkle in the bottom-right corner that appears on content made with the Nano Banana and Omni models. The same control is coming to Flow, Google’s AI video editor, and it also covers music made with Lyria, Google’s music generation model.
The announcement came from Josh Woodward, the vice president of Google Labs, Gemini, and AI Studio. He framed the change as a balance between creative control and safety. Google is not removing provenance markers, though. Invisible SynthID watermarks and C2PA metadata stay embedded in the background even when the visible mark is off, so you can still verify whether a piece of content is AI-generated by asking Gemini or Search.
The invisible marks stay
SynthID is Google’s invisible watermarking layer, and it stays embedded in the background even when the visible mark is off. C2PA metadata rides along too, which is the part verification tools and platforms read. Google is also open sourcing Credentio, a library that lets developers embed local validation in their own apps. The practical result is that the provenance layer gets stronger even as the visible badge becomes optional. The check just moves from a glance to a deliberate action, which is the real change for anyone trying to spot AI content in the wild.
What this means for spotting AI content
The visible sparkle was the easy tell. Once the Media watermark toggle lands for you, that quick check is gone, and spotting a Gemini image at a glance gets harder. The invisible layer remains, and Google says you can still ask whether something was created with its tools.
This puts Google in line with the rest of the field. Other AI image generators like OpenAI and Meta don’t apply visible watermarks, and OpenAI relies on SynthID and C2PA instead. Anthropic this week started applying invisible watermarks to AI-generated text and images. Google’s move makes the visible badge optional while keeping the detection layer underneath. The end result is that the badge becomes a choice for creators, and the burden of checking shifts to anyone who actually cares whether something is AI-made.
For regular users the trade-off is real. A visible stamp is honest but ugly, and plenty of people already crop it out before sharing. Making it optional acknowledges that the badge was never the real safety mechanism, and the invisible layer is the one that actually survives edits and screenshots. The change reflects an evolving approach to labeling AI-generated media, because visible watermarks can make content less useful for professional and creative work, and many users have already been editing them out anyway.
Where the Media watermark toggle won’t show up
Google plans to bring the Media watermark setting to Search as well, but the rollout has limits. The toggle won’t launch in countries that require visible watermarks by law. The EU’s AI labeling rules, which went live earlier this month, are exactly the kind of requirement Google is talking about. When a jurisdiction mandates the visible mark, the setting stays off.
My take
The practical change is small and the strategic one is bigger. For anyone who makes images with Gemini, killing the corner badge removes a real annoyance, and the invisible marks mean the provenance story mostly survives. The reason this matters is that visible watermarks were always a weak signal anyway: they crop out, they get covered, they get removed by anyone with an editor. Google is admitting that and moving the trust to the layer that actually holds up.
The Media watermark switch is also a rare consumer-facing win from a company that spends most of its energy on model benchmarks. Google says the setting is rolling out over the coming days, with Search support planned, so most people will hit it sooner than later. Google’s Gemini 3.6 Flash launch in July showed the same pattern: fast iteration on what developers and users actually ask for. I would still keep asking for the invisible verification tools, because that is the part that keeps everyone honest once the sparkle is gone.
If you generate a lot of images or video, check the settings menu in the next few days and flip the Media watermark option to match how you actually share your work. Just remember that turning the badge off does not make the output anonymous. The SynthID mark and C2PA metadata are still in there, and that is a feature, not a catch.
