If you get a good chuckle out of unrealistic videos generated by contemporary AI models, you might want to get your laughs in now while you still can. The debut of Google DeepMind‘s Veo 3 showed the world that generative AI programs can now make videos that are difficult to distinguish from human content. Brace yourselves: The internet is about to get more confusing than it already is.
Google introduced Veo 3 on May 20 during its annual I/O developer conference. The new model improves on its predecessor by adding sound effects, background noise, dialogue, and other forms of audio to the videos it generates.
“For the first time, we’re emerging from the silent era of video generation,” said DeepMind CEO Demis Hassabis during a press briefing. Hassabis noted that users can provide “a prompt describing characters and an environment, and suggest dialogue with a description of how you want it to sound.”
Subscribe to get the latest creator news
The audio enhancements Veo 3 offers for generative AI are potentially powerful, but the improvements it brings to video quality are arguably its most stunning features. A post by X user Min Choi showed a snippet of a man-on-the-street interview generated by Veo 3. If you didn’t know ahead of time that this was made by AI, would you be able to suss it out?
Google Veo 3 realism just broke the Internet yesterday.
This is 100% AI
10 wild examples:
1. Street interview that never happened pic.twitter.com/qdxZVhOO3G
— Min Choi (@minchoi) May 22, 2025
The examples Choi provides show that AI-generated video has come a long way since OpenAI’s Sora arrived in February 2024 — and that model was already a revelation at the time of its release. Veo 2 was powerful enough to be integrated into YouTube features like Dream Screen, and as we can see, its successor is far more advanced.
The rapid progression of generative AI technology validates individuals who claim that we’re heading toward a “dead internet” populated solely by AI-generated videos and comments. If unrealistic genAI videos are already popular enough to rank among YouTube’s most-watched fare, how many views will realistic genAI videos get?
To make matters worse for (human) creators, there’s a good chance Veo 3’s training material includes a significant number of YouTube videos — though Google has been cagey about that subject. If a large language model knows more about viral YouTube content than the creators of those videos, how will humans keep up with the AI revolution?
The answer to that question will involve a lot of vigilance from platforms, creators, and viewers. YouTube has taken strides to ensure that generative AI videos are properly labeled, and all content made with Veo 3 will be stamped with Google’s SynthID watermarks.
Even if information about AI-generated videos is available, it doesn’t mean that viewers will use it. There has been a lot of talk lately regarding media literacy, and with advanced generative AI tools staring us in the face, it’s more important than ever to teach kids and teens how to spot fakery on the internet.
Veo 3 can’t be written off as a passing tech fad. Hollywood directors are already incorporating DeepMind products into their productions, with auteur filmmaker Darren Aronofsky serving as one of the latest examples. Internet users who ignore the potential of the latest models will have a harder time identifying the work of those programs in the wild.
If you want to see for yourself what Veo 3 is capable of, you’ll need a Google AI Ultra subscription, which will run you $249.99 per month. Alternatively, the new model is available on Google’s Vertex AI enterprise platform.
