Google is giving Gemini and Flow users a choice they didn’t have before: whether AI-generated images, video, and audio carry a visible watermark at all. The invisible tracking layer underneath stays on regardless, which is the detail most coverage of this update is glossing over.
The change matters less as a design tweak and more as a signal of where AI content provenance is headed — toward invisible, machine-readable proof rather than a logo stamped in the corner of an image.
What Google actually changed
Josh Woodward, who leads Google Labs, the Gemini app, and AI Studio, announced the update on X. A new setting called Media Watermark lets people turn the visible watermark on Gemini and Flow output on or off, on a going-forward basis.
The toggle spans three of Google’s generative models: Nano Banana for images, Omni for video, and Lyria for music. Google is rolling it out across Gemini and Flow over the next several days, with support for Google Search listed as coming next, though no date has been set for that rollout.
Anyone who wants to change the setting can find it under Settings > Media Watermark inside Gemini. Flip it and every piece of content generated after that point follows the new preference — it does not retroactively strip watermarks from files already created.
Woodward framed the decision as a tradeoff: “We’re striking a balance here between creative control and safety,” he wrote, adding that the invisible layer — SynthID watermarking and C2PA metadata — keeps running underneath either way.
The part that doesn’t change: the invisible layer
This is the detail worth sitting with. Removing the visible watermark is a cosmetic change. Every file Gemini and Flow produce still carries a SynthID watermark embedded in the pixel or audio data itself, plus C2PA Content Credentials metadata recording that the file was AI-generated. Neither of those depends on whether the visible mark is showing.
Both layers remain checkable after the fact. Anyone can ask Gemini directly to verify whether an image, video, or audio file carries a SynthID watermark, and Google’s “About this image” feature, AI Mode, and Lens all perform the same kind of check independent of the visible watermark setting. In practice, that means turning off the visible mark changes what a casual viewer sees, not what a determined checker can find.
Google reinforced that infrastructure this week by open-sourcing Credentio, a C++ library that lets developers verify C2PA credentials inside their own applications without sending the file to Google’s servers first. Google says the library already runs inside close to 40 of its own C2PA-conformant products — a sign the company is trying to make provenance-checking something any developer can build into their own tools, not just a feature locked inside Google’s apps.
Where the toggle won’t show up
The option isn’t universal. Woodward said the setting won’t appear in countries where a visible watermark is a legal requirement, though Google hasn’t published a specific country list for the new Gemini toggle yet.
The closest available signal is Flow’s existing help documentation for video, which already restricts the visible watermark in India, South Korea, and Vietnam. Given that Flow and Gemini share the same underlying models, it’s a reasonable bet the Gemini toggle will follow the same regional restrictions once Google publishes the full list. At the time of writing, Google’s help pages for the new Gemini setting hadn’t gone live, so the exact scope is still unconfirmed.
Why this is bigger than a settings toggle
Visible watermarks have always done two different jobs at once: they signal to a casual viewer that content is AI-made, and they act as a copy-deterrent against people passing the content off as something else. Making the visible mark optional splits those jobs apart. The deterrent function largely disappears — nothing stops someone from screenshotting or downloading unwatermarked output and presenting it without disclosure. The disclosure function, though, doesn’t actually go away; it just moves entirely into infrastructure most people never interact with directly.
That’s a meaningful shift for anyone publishing AI-assisted content, including agencies and in-house teams producing images, video, or audio for client work. A file with no visible watermark can still be traced back to Gemini by anyone who runs it through a verification tool, which means the assumption that “no watermark” equals “no record” doesn’t hold. Platforms, publishers, and eventually regulators building AI-disclosure requirements around C2PA metadata rather than visible marks will have more durable enforcement than a stamp anyone can crop out.
It also puts Google in a slightly different position than competitors that rely primarily on visible marks with no equivalent invisible layer. SynthID and C2PA together give Google a provenance system that survives the exact evasion — cropping, screenshotting, re-encoding — that defeats a simple visible watermark. Whether that holds up against more aggressive tampering (heavy compression, format conversion, adversarial editing) is a separate and still-open question, but the baseline is stronger than a corner-of-the-image logo.
What this means if you use Gemini or Flow for content production
For teams generating marketing images, video assets, or audio with Gemini and Flow, the practical takeaway is narrow but useful. Turning off the visible watermark can make output look cleaner for client presentations or ad creative, but it doesn’t buy plausible deniability about how the asset was made — the metadata trail persists regardless of the setting. Teams that need to prove provenance for legal, compliance, or platform-disclosure reasons should keep relying on the SynthID/C2PA verification tools rather than the visible watermark’s presence or absence, since the latter is now purely a user preference rather than a reliable indicator either way.
The rollout to Search is the piece to watch next. If Google extends watermark controls into Search results and AI Overviews, it starts to touch how AI-generated visuals surface in organic search — an area Google has already been expanding aggressively with Gemini models inside AI Mode. Combined with Gemini’s continued growth — the app recently passed a billion monthly users — the scale at which this provenance system operates is only going to increase.
The trust question underneath it all
Watermark policy is really a trust policy wearing a technical costume. As AI systems take on more autonomous capability — Gemini can now control a user’s computer directly, and the ecosystem around prompt-based workflows keeps expanding through features like reusable Gemini prompt workflows in Chrome — the question of what’s verifiably AI-made and what isn’t only gets more consequential. A visible watermark was always a blunt, easily defeated signal. An invisible, cryptographically-tied metadata layer that survives cropping and screenshotting is a genuinely different kind of commitment, and it’s the one Google chose to keep non-optional.
Frequently asked questions
Does turning off the visible watermark make Gemini content untraceable?
No. SynthID watermarking and C2PA Content Credentials metadata are embedded in every file Gemini and Flow produce regardless of the visible watermark setting. Anyone can still verify a file’s origin using Gemini directly or tools like “About this image,” AI Mode, and Lens.
Which Gemini models does the new setting cover?
Three: Nano Banana for images, Omni for video, and Lyria for music. The toggle lives under Settings > Media Watermark in Gemini and applies to future content only.
Will the toggle be available everywhere?
No. Countries with legal requirements for visible AI watermarks won’t get the option. Google hasn’t published a full list for the Gemini toggle, but Flow’s existing restrictions cover India, South Korea, and Vietnam for video.
Is Google Search getting this feature too?
Google has said support for the watermark setting is coming to Search, but has not given a timeline or details on how it will function there.
What is Credentio?
Credentio is a C++ library Google open-sourced this week that lets developers verify C2PA credentials inside their own applications without routing files through Google’s servers. Google says it’s already used in close to 40 of its own C2PA-conformant products.