If the internet wasn’t already full of misinformation and general filler content, Veo 3 is going to take it to an entirely new level. Google’s newly announced Veo 3 video and audio generation engine is already being used to create terrifyingly realistic videos that would fool just about anyone.
The new Veo 3 engine was announced at Google I/O, along with a slew of other upgrades to Google’s AI models. The big improvement to the Veo engine is the addition of synchronized audio. It turns out, audio was the one thing that separated obviously AI content from real video clips with actual human beings.
In addition, the video generation aspect is better, too, making it all the more difficult to differentiate generated content from organically created content. Frames are crisper and filled with further detail, which only adds to the realism that’s now so hard to pick apart.
The additional tools Google has given users add fuel to the fire. The recently announced Flow tool in Google Labs allows users to create clips and seamlessly merge them like a virtual storyboard. If done correctly, AI clips become movies with continuity and exposition, if the user wishes. Even though single clips are limited to around eight seconds, Flow allows for much longer videos with stitched scenes.
Even though Veo 3 is hidden in Google’s AI Ultra plan, users are already starting to fill social media with clips they’ve created. One post on Reddit comes from a user who used Veo 3 to create a one-minute video clip that resembles something we’d see from the guys at Electrek during a car show.
The clip showcases several interviewees, all with incredibly detailed synchronized audio and individual personalities. Each person has a unique look and has a style of talking that makes one think it could be a real car show.
Again, these are not real people; they are whatever Veo 3 dreamed up. The image seen and the audio generated are entirely synthetic, even though the audio sounds so completely real. There are a couple of telling signs in this one, like the repeatedly appearing dad with child, or the coffee cup that sounds like glass when the speaker puts it down.
As nitpicky as those details are, that’s what it now takes to decipher whether a video is AI or not.
Another Veo 3 clip portrays a YouTuber reacting to Minecraft gameplay and explaining how to play. This one is a little easier to brush off as false for a couple of reasons. Mainly, the Minecraft gameplay looks real but poses a couple of far-fetched features, like how breaking wood with your fists adds bread to your inventory. That’s a famous mechanic and not AI-dictated.
Veo 3 even generates the green-screen haze around the person’s hair correctly, which is wild detailing to incorporate.
The issue with these gameplay videos is that they look incredibly accurate, and it takes pixel-peeping to determine whether they’re generated or not. Take a Fortnite gameplay video that was generated with Veo 3, for example. The only real way to determine that the entire video is fake is the lack of UI elements when the player wins the game. Otherwise, you’d think it was a new YouTuber’s recording setup.
The above clip comes from the simplest prompt, too:
Streamer getting a victory royale with just his pickaxe
As incredibly impressive as these generated videos with synched audio are, it’s important to predict the sort of content we might see on the internet going forward. These clips could fool most people, without a doubt, and the propensity for content creators with the opportunity to develop misleading information with these tools is high. This is the step where AI merges with reality.
More on Gemini:
- Flow is Google’s new AI video editing suite powered by Imagen 4 and Veo 3
- Google AI wants to do all of your shopping with instant checkout and AI virtual try-on
FTC: We use income earning auto affiliate links. More.
Comments