Even at first glance, there is something good on the street. The white sheet, under which it is, is a bit too neat, and the movements of the officers are completely free of purpose. “We have to clean the street,” says one of them with a robust gesture of hands, although its lips do not move. This is AI, okay. But here is Kicker: My prompt contained no dialogue.
VEO 3, a novel Google video generation model, added that it was all alone. Over the past 24 hours, I have created a dozen clips depicting press reports, disasters and a stupid cartoon with a convincing sound – some of them were invented. It’s more than a bit scary and much more sophisticated than I imagined. And although I do not think that it will lead us to disinformation, that VEO 3 strikes me as an absolute AI Slop Machine machine.
Google Present that I see 3 On I/O this week, emphasizing its most significant novel ability: generating sound for your AI movie. “We enter the new era of creation,” explained the Vice President of Google from Gemini, Josh Woodward, calling it “extremely realistic.” I was not completely sold, but a few days later VEO 3 I generated a video with an information anchor announcing a fire in space needle. A basic text prompt was enough, a few minutes and a costly subscription to the Ultra AI Google plan. Do you know what? Woodward did not exaggerate. It’s realistic as hell.
I tried the monitor about anchors after the news after seeing what Alejandra Caraballo, a clinical instructor at Cyberlaw Clinic Harvard Law School, was able to produce. One of her clips He has an information anchor announcing the death of the American Secretary of Defense Pete HegeSth. He is not dead, but the clip is extremely convincing. Post, including the series of films with Characters generated by AI protesting for the hints used to create them Has 50,000 meals per reddit. The scenes include disasters, a woman in a hospital bed with a respiratory tube and a character threatened by a sight – all with spoken dialogue and realistic background sounds. Real carefree things!
Maybe I’m naive, but I’m not having fun with VEO 3 quite As worried as at the beginning. At the beginning there are obvious handrails. You can’t monitor this to create a video biden tripping and falling. You cannot announce that the anchor will announce the murder of the president, and even generate a video from the Director General of Technology Technology with a T -shirt and chain, while the dollar bills fall around him. This is the beginning.
Having said that, you can generate disturbing shit. Without a clever bypass, I prompted VEO 3 to create a film with a space needle on fire. Starting from my own photo, Mount Rainier, I generated his video exploding with smoke and lava. In conjunction with the information of the information anchor announcing the aforementioned catastrophe, I see how straightforward it is easily harm to mischief with this tool.
Here is a better message: it doesn’t look like a ready -made depth machine. I gave him some photos and asked to generate a movie with a specific dialogue and it would not be consistent. I also asked him to bring a pair of giant shoes for a photo and ask them from the stage; He managed one shoe stinging on the pavement with comic sounds of crunching in the background.
I had a easier time to generate movies when my hints were less detailed, in this way I confirmed something My colleague Andrew Marino pointed out: VEO 3 is perfect in creating the type of the lowest YouTube content denominator targeted at children.
If you have never been subjected to an endless garbage cavity on YouTube Kids, let me enlighten you. Imagine watching the worst rendering 3D monster truck riding on the ramp, landing in a ladle with colorful paint. Next to him, another monstrous truck enters another ramp to another vat with paint – this time a different color. Now watch it again. And again. And again. On YouTube there are hours designed to hypnotize youthful children. These films are usually harmless, simply empty calories designed to get views that make Cochelon look Citizen kane. In about 10 minutes with VEO 3 I dropped the clip after the same basic formula – along with cheerful music in the background. But the clip, which is even more disturbing to me, are two cartoon cats on the pier.
I thought it would be humorous if the cats complain about each other that the fish did not bite. In just a few minutes I had a clip with two cats and some dialogue generated by AI, which I never wrote. If it is so straightforward to make a 10-second clip, stretching it to a seven-minute video on YouTube would be inconsequential. In its current form, the clips return to VEO 2 when you try to expand them to longer scenes, which removes sound. But the way Google moves these tools mercilessly, I can’t imagine that it will be long before you can edit the full length of the VEO 3 video function.
To be straightforward, I wonder if this type of video generated by AI is a function, not a mistake. Google showed us fancy video generated by AI real filmmakers, including Eliza McNitt, who works with Darren Aronofsky over a novel film with some AI generated elements. And it is clear that AI video can be an engaging tool in the right hands. But I think that what we will most likely see is the spread of a dull image that AI is so good to generate – this time in stereo.