.png){kind=logo}
Google makes AI watermarks optional across images, videos, and music created with AI.
Invisible SynthID remains active, preserving machine-readable identification after labels disappear from AI content.
Journalists need provenance tools, verification systems, and source checks to identify synthetic media.
Google has made a notable change to the way it labels AI-generated media. Users can now turn off the visible watermark on content created with Google’s AI tools, including images, videos, and music. The move gives creators greater control over how their work appears, but also raises questions about how easily people will be able to identify AI-generated content online.
The company, however, is not abandoning AI-content identification. Invisible watermarking and provenance technologies will continue to operate even when the visible mark is removed.
Google is introducing a “Media watermark” setting that allows users to remove the visible sparkle-style watermark from AI-generated content. The option applies to content created through tools and models including Gemini and Flow, with Google’s Nano Banana, Omni and Lyria models among those covered by the change. The feature is rolling out across Gemini and Flow, with support for Search expected soon.
The visible watermark had served as an immediate signal to viewers that an image, video, or song was generated using AI. Google’s latest move makes that signal optional.
The important distinction is that Google is not switching off AI-content identification altogether. Even after the visible mark is removed, Google says invisible SynthID watermarks and C2PA metadata will continue to be used for transparency. SynthID is designed to embed an imperceptible signal in AI-generated content, enabling detection by compatible systems.
This means Google is effectively moving towards a two-layer system: a visible label that users can choose to remove, and an invisible identification layer that remains embedded in the content. Google has said users can still use Gemini or Search to check whether an image was AI-generated.
To some extent, it could. For an ordinary user scrolling through social media, the visible watermark is the easiest way to understand that an image or video has been created using AI. Removing that visual cue means viewers may no longer immediately know where the content came from.
But there is another side to the argument. Visible watermarks can interfere with creative and professional work, particularly when generated images, videos, or music are intended for presentation or commercial use. Google has described the change as an attempt to strike a balance between creative control and safety.
There is also a practical limitation: visible watermarks can already be cropped or edited out. Making them mandatory does not necessarily guarantee that an AI-generated file will remain visibly labeled after it moves across the internet.
Also Read: Google Gemini Lets Users Hide Visible AI Watermarks
For journalists and fact-checkers, the change makes one point particularly important: the absence of a visible watermark cannot be treated as proof that a human created content.
Newsrooms will increasingly need to look beyond what is visible in an image or video. Provenance information, C2PA credentials, SynthID detection, reverse image searches, source verification, and independent confirmation can all be important parts of the verification process.
Google has also been expanding tools intended to help users understand where digital content came from. The company has been working with industry standards and technologies designed to improve the identification of AI-generated media.
Also Read: Google Maps Trip Reveals 390-Million-Year-Old Meteorite Crater
Google’s decision comes as AI transparency becomes a bigger regulatory and industry issue. The European Union’s AI Act includes transparency obligations covering AI-generated content, increasing pressure on technology companies to make synthetic media identifiable.
Other AI companies are also taking different approaches. Anthropic has introduced imperceptible watermarking for AI-generated text and C2PA provenance for images, showing how the industry is increasingly moving towards machine-readable identification rather than relying only on visible labels.
Google’s decision therefore does not mark the end of AI watermarking. Instead, it changes what users see. The visible warning can now disappear, but the digital trail is intended to remain. For consumers, that may mean fewer obvious clues. For journalists, platforms and fact-checkers, it means provenance and verification systems will become even more important in determining whether digital content is AI-generated.
The larger question is whether invisible systems such as SynthID and C2PA can become widespread, reliable, and simple enough for ordinary users. If they do, AI transparency could gradually shift from ‘look for the watermark’ to ‘verify the content’s provenance.’
Why this Matters?
Google’s move gives creators more control but removes an obvious AI signal. As synthetic media spreads, journalists and users will increasingly depend on invisible watermarks and provenance tools.
Why is Google making AI watermarks optional?
Google wants to give creators greater control while retaining invisible identification through SynthID and other provenance technologies.
Can users completely remove Google’s AI watermark?
Users can remove the visible watermark, but invisible SynthID identification remains embedded in supported AI-generated content.
What is Google SynthID?
SynthID is Google’s invisible watermarking technology designed to identify AI-generated images, videos, audio and text.
Does removing the watermark make AI content harder to verify?
Yes, visible identification becomes harder, making provenance tools, SynthID detection and source verification increasingly important for journalists and users.
What does Google’s move mean for journalists?
Journalists should not rely on visible watermarks alone and should use provenance, metadata, reverse searches and independent source verification.
