Last week, Google discreetly announced it would introduce a visible watermark for AI-generated videos utilizing its new Veo 3 model. If you pay close attention while browsing through your social media feeds, you may catch a glimpse of it. The watermark is present in videos issued by Google to advertise Veo 3 in the UK and various other regions.
This update was shared in an X thread by Josh Woodward, Vice President at Google Labs and Google Gemini. Woodward stated that the watermark is included on all Veo videos except those created in Google’s Flow tool by users subscribing to a Google AI Ultra plan. This newly introduced watermark supplements the invisible SynthID watermark already integrated into all of Google’s AI-generated materials, along with a SynthID detector that is currently undergoing testing but is not yet widely accessible.
The visible watermark “is a first step as we strive to make our SynthID Detector available to a broader audience in parallel,” remarked Josh Woodward, VP of Google Labs and Gemini in his X post. Since the unveiling of Veo 3 at Google I/O 2025, the AI video model has attracted attention for its remarkably lifelike videos, which include realistic audio and dialogue. The videos online aren’t merely fantastical portrayals of animals behaving like humans, though there is certainly no shortage of that. Veo 3 has also been employed to create more ordinary clips, such as street interviews, influencer advertisements, fabricated news segments, and unboxing videos.
Upon close inspection, you can identify telltale signs of AI, such as overly-smooth skin and incorrect artifacts in the background. However, if you’re scrolling casually, you might not think to verify if the emotional support kangaroo clutching a plane ticket is genuine or artificial. The deception caused by an AI-generated kangaroo is relatively benign, but the extensive availability and realism of Veo 3 raise new concerns about misinformation, according to AI specialists consulted by Mashable.
The new watermark should theoretically mitigate these risks. The issue, however, is that the visible watermark lacks visibility. In a video Mashable created using Veo 3, a “Veo” watermark in light white appears in the bottom right corner. Can you spot it?
“This small watermark is unlikely to be noticeable to most consumers swiftly scrolling through their social media feeds,” stated digital forensics expert Hany Farid. It took us several seconds to locate it, and we were actively searching for it. Unless users are aware to look for the watermark, they may overlook it, particularly on mobile devices.
A Google representative informed Mashable via email, “We are dedicated to developing AI in a responsible manner and have established clear policies to safeguard users from harm and regulate the use of our AI tools. Any content produced with Google AI has a SynthID watermark integrated, and we also append a visible watermark to Veo videos.”
“People are accustomed to prominent watermarks, like those from Getty Images, but this one is quite small,” commented Negar Kamali, a researcher analyzing individuals’ ability to identify AI-generated content at Kellogg School of Management. “Either the watermark needs to be more conspicuous, or platforms hosting images could include a notice next to the image — something like ‘Check for a watermark to confirm if the image is AI-generated,’” suggested Kamali. “With time, people could learn to search for it.”
Nonetheless, visible watermarks are not a flawless solution. Both Farid and Kamali pointed out that videos with watermarks can easily be cropped or altered. “None of these minor — visible — watermarks in images or videos are adequate because they can be easily removed,” remarked Farid, who is also a professor at UC Berkeley School of Information.
However, he highlighted that Google’s SynthID invisible watermark “is quite durable and challenging to eliminate.” Farid added, “The drawback is that the typical user can’t perceive this [SynthID watermark] without a watermark reader, so the current objective is to simplify it for consumers to identify whether a piece of content contains this type of watermark.”