Google’s visible watermark on VEO 3 videos is actually invisible

Last week, Google quietly announced that it would add visible watermarks to AI-generated videos made using its new VEO 3 model.

And, if you’re very careful while scrolling through social feeds, you might be able to see it.

Watermarks can be seen in Google-released videos to promote VEO 3 launches in the UK and other countries.

Google announced changes in X-Threads by Google Labs and Google Gemini Vice President Josh Woodward.

According to a post by Woodward, the company has added watermarks to all VEO videos, except for videos generated by users through Google’s streaming tools with Google AI Ultra Plan. In addition to all the contents generated by Google AI and the invisible synthetic watermarks in synthetic detectors, the new watermarks are embedded, which recently pushed to early testers but are not widely available yet.

This tweet is currently unavailable. It may be loading or being deleted.

The visible watermark “is the first step because we strive to make more people available in parallel,” said Josh Woodward, vice president of Google Labs and Gemini, in his X post.

Within the weeks after Google introduced WEO 3 on Google I/O 2025, the new AI video model has caught the attention of incredible realistic videos, especially since it can also generate realistic audio and conversations. The videos posted online are not just fantasy renderings of animals that are portrayed by humans, although there are many. VEO 3 is also used to generate more mundane clips, including street interviews, influential ads, fake news clips and unboxing videos.

This tweet is currently unavailable. It may be loading or being deleted.

If you look closely, you can spot obvious signs of AI, such as excessively smooth skin and wrong artifacts in the background. But if you are passively doomed to roll, you probably won’t consider double checking whether the emotional support of the Kangaroo casually holding the ticket is real or fake. People who are deceived by AI-generated kangaroos are a relatively harmless example. But, according to AI experts interviewed by Mashable, the widespread availability and realism of Weo 3 present new risks to spreading false information.

In theory, new watermarks should reduce these risks. The only problem is that the visible watermark is not That Visible. In the video mashup generated using VEO 3, you can see a “veo” watermark in the lower right corner of the video. See it?

Mashable generated VEO 3 videos include new watermarks.
Image source: Screenshot: Mashable

How is it now?

Google’s VEO watermark.
Image source: Screenshot: Mashable

“This little watermark is unlikely for most consumers who feed social media with breakthrough sandwich,” said Hany Farid, a digital forensics expert. Indeed, it took us seconds to find it, and we were looking for it. Unless users know to look for a watermark, they may not see it, especially when viewing content on their mobile devices.

A Google spokesperson told Mashable via email: “We are committed to developing AI responsibly and we have clear policies to protect users from damage and manage the use of AI tools. Any content generated with Google AI is embedded with SynthiD WaterMark, and we also add a visible watermark to VEO VEOOS.”

“People are familiar with prominent watermarks like Getty Images, but that’s small,” Negar Kamali said. “So the watermark needs to be more eye-catching, or the platform hosting the image may include notes next to the image – similar to ‘checking the watermark to verify that the image is generated for AI,'” Kamali said. “Over time, people can learn to look for it.”

However, the visible watermark is not a perfect remedy. Both Farid and Kamali tell us that videos with watermarks are easy to be cropped or edited. “None of these small images or videos have enough watermarks because they are easy to delete,” said Farid, who is also a professor at the UC Berkeley School of Information.

However, he noted that Google’s synthetic watermark is “very elastic and difficult to clear”. Farid added: “The disadvantage is that the average user cannot see it.” [SynthID watermark] There is no watermark reader, so the goal now is to make it easier for consumers to know if a piece of content contains such a watermark. ”