Google lets you remove the visible watermark from AI images, videos and songs
The visible stamp is now optional. The hidden watermark that proves an image is AI-generated stays put.
Key points
- Google announced on 14 August 2026 that users can now turn off the visible watermark on AI-generated images, videos and songs.
- SynthID, the invisible watermark, and C2PA metadata remain active regardless of the setting.
- The toggle covers Nano Banana (images), Omni (video) and Lyria (music) inside the Gemini app and the video editor Flow, with Search support coming soon.
- Google is open-sourcing a library called Credentio so developers can add AI-content verification directly inside their own apps.
If you have ever created an image or song with Google's AI tools and found the visible stamp awkward for professional work or a design project, you can now turn it off. Google announced on Friday that the visible watermark on AI-generated content is optional across its main creative AI tools.
A stamp printed across an AI image can make it harder to use in a presentation or a piece of marketing. Google is responding to that friction.
Does this mean AI-generated content will be harder to detect?
No. The visible watermark was always the easy part to spot. The protections that actually matter stay on.
Google's Josh Woodward, the company's vice-president for Gemini, confirmed in a post that SynthID and C2PA metadata remain active. SynthID is an invisible watermark baked into the image or audio file itself, below the level a human eye or ear can detect. C2PA metadata is background data attached to a file recording where and how it was created. Neither is affected by the new toggle.
"We're striking a balance here between creative control and safety," Woodward wrote. "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."
The move follows Anthropic's decision to embed invisible markers into every Claude response, which we reported on 11 August in "Anthropic Will Hide Secret Watermarks in All Claude Text and Images". Where Anthropic added hidden signals and kept them mandatory, Google is making the visible layer removable while holding the hidden layer firm.
Which tools and models does this cover?
The setting covers Nano Banana (image generation), Omni (video) and Lyria (music). It will appear inside the Gemini app and Flow, Google's AI video editor. Support inside Google Search is coming soon.
Once the feature reaches your account, find the toggle at Settings, then Media Watermark.
| Model | Content type | Where to find the toggle |
|---|---|---|
| Nano Banana | Images | Gemini app |
| Omni | Videos | Gemini app, Flow |
| Lyria | Songs | Gemini app |
Google is rolling the feature out over the next few days, so not every account will see it immediately.
Google is also open-sourcing Credentio. Developers can use it to add local AI-content verification directly inside their own products, without relying on Google's servers to do the checking. That matters: it means third-party apps could flag AI-generated content even without an internet connection to query Google.
Common questions
If I turn off the visible watermark, can people still tell my image was AI-made?
Yes, if they use the right tool. The hidden SynthID signal stays in the file. Both Gemini and Google Search can read it to flag the content as AI-generated.
Do I need to do anything to keep my content labelled?
Nothing changes by default. The invisible watermark and C2PA metadata stay on unless Google changes that policy in future. The new setting only controls the visible stamp on screen.
