Matt Wolfe breaks down three telltale signs of AI video: unrealistic physical actions like jumping or throwing, distorted human shapes from models trained on flawed AI footage, and awkward pauses between sentences. He uses examples from Sora 2 and VO 3.1 to show how fast the tech is improving.
Here are the three things that even the best Gen AI platforms still struggle with I guess AI video platforms and I have something in common. We're both not physicists. Actions like jumping, swinging, falling, or throwing will usually take a lot of prompting to get a realistic looking AI video. We've mostly solved the six fingers issue, but AI video is still creating some crazy shapes out of people. I think this is still an issue because of the AI training loop that's happening where the AI is training itself on other AI videos that had the same issues, so it's not actually getting better and can lead to model collapse. This is also why a lot of AI generated content is actually not of people, but of things like animals or objects with less moving parts. If you're trying to make a longer video with multiple parts, you might notice some awkward pauses in between sentences.
Yes, nailed it. Two thumbs up. I guess this is where the AI is thinking or something. But this is also the easiest one to fix since you can just cut out those pauses pretty easily. AI generated video is getting better every day, so a lot of the past tips and tricks for spotting it are kind of outdated now. But almost all of the video examples that you just saw were made with the two best gen AI platforms today, Sora 2 and VO 3.1, and they both still struggled with the three things that I mentioned. So whether you're trying to avoid AI video altogether, or you're trying to make your own AI videos, spotting these weak points will really give you an edge.
AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know
