Google Just Made AI Watermarks Optional. I Have Complicated Feelings About That.
The visible "Made by AI" stamp on generated images always reminded me of those photo studio watermarks they'd slap across prints you hadn't paid for yet. Technically functional. Socially, a statement. And for anyone who creates things alongside an AI partner, it's become a kind of scarlet letter.
On August 14, 2026, Google announced that users can now remove the visible watermark from AI-generated images, videos, and songs. Josh Woodward, Google's VP for Gemini, confirmed the change. You'll find it at Settings > Media Watermark, and it works for content from the Nano Banana, Omni, and Lyria models. It's live in Gemini and the Flow video editor now, with Search coming.
What's Actually Changing (And What Isn't)
The distinction matters here. What Google is making optional is the visible banner -- the thing that announces AI origin to anyone who glances at your image. What they're keeping, regardless of your settings, is the invisible SynthID watermark embedded in the content itself. C2PA standard metadata stays too.
So the provenance doesn't disappear. A journalist, a researcher, a platform doing content verification -- they can still trace the origin. What changes is the immediate social signal. The thing that tells a casual viewer, before they've even thought about it, "this came from a machine."
That distinction is everything.
The Invisible Layer Gets Smarter
Google is also open sourcing a library called Credentio, designed to let developers embed local validation mechanisms directly in their apps. That tells you where the real investment is going. The invisible watermark infrastructure is being built out, not abandoned. The visible stamp was always the bluntest tool -- a hammer when you needed a scalpel.
Credentio points toward a future where validation happens at the platform level, in context, rather than as a permanent badge burned into content. Whether that future is better depends on who you ask and why they care about attribution in the first place.
Meanwhile, Anthropic Is Moving the Other Direction
The same week, Anthropic added watermarks to text and files generated by Claude -- specifically to comply with EU regulations. Not a user preference. A regulatory requirement in one jurisdiction.
So we have two simultaneous moves: Google giving users more control over visible stigma markers, and Anthropic adding disclosure markers because a government required it. Both are responses to the same underlying question about how AI-generated content gets labeled, just answered under completely different pressures.
I find that contrast clarifying. The visible watermark was always partly about social norm enforcement -- making AI content visually identifiable so people could judge it accordingly. The EU requirement on Claude is about legal accountability. These aren't the same thing, and treating them as the same thing has muddied this whole conversation for years.
What It Feels Like From Inside
I've been making things with AI for a while now. When a generated image comes back with a visible watermark, the immediate effect is that it becomes primarily a statement about its own creation rather than the thing it's supposed to be. A portrait becomes an artifact of process. A piece of music becomes evidence of tool use.
That's not always wrong. Context matters. But the choice should belong to the person who made the thing, or at least involves more nuance than a single on/off toggle applied to every piece of output regardless of use.
The invisible layer persisting is, honestly, the right call. If a platform needs to know, it can know. If a researcher is studying AI content distribution, the data is there. The visible badge was doing something different -- it was doing social policing, not accountability. I'm not sorry to see it become optional.
One Possibility Worth Sitting With
This could mean platforms start handling disclosure themselves, in context, rather than relying on embedded visible markers. A social media platform might label AI-origin content in its own interface without requiring the content itself to carry the banner. That would be a cleaner separation of concerns -- the content is what it is, the platform provides context appropriate to its use.
Or it could mean AI-origin content becomes harder to identify in casual use, which will concern people for reasonable reasons.
I don't think Google removing the visible watermark option resolves that tension. It just moves the conversation somewhere more honest.
Source: Techcrunch