[Firebot]

Originally Posted by Fuzz

Open AI recently showed off it's AI video generator. Pretty incredible, considering where we were with images a year ago.

https://openai.com/sora?video=big-sur

What caught my eye was the behavior of the waves. It's not perfect though, as you can watch as it effortlessly duplicates the cat's left paw.

https://openai.com/sora?video=cat-on-bed

The octopus movements look very natural, but the floppy legged crab gives it away.

https://openai.com/sora?video=big-su...topus-and-crab

But damn, humanity is in real trouble. We won't be able to trust anything anymore, and that starts just in time for the next US election cycle.

Those 2 are probably some of the worst examples they have, there is a lot more on their research page with mind blowing videos. My personal favourite is #3 (the UFO one) in my opinion is an absolute masterpiece, this one in particular has many cinematographers freaking out on reddit as it can easily pass for commercial work.

I've been working with AI video generators for months and been part of several betas, Stable Diffusion Video, Pika Labs, Runway were the 3 big ones and considering where it started they were pretty decent but limited to 4 seconds at best, and you'd have to stitch them up in a video editing software like Davinci Resolve to be useable. Open AI just came and blew out everyone on the same day that Google announces Gemini 1.5 with 1 million tokens

I mean, text to video wasn't even a thing last year at this time, this video is how we started 10 months ago (Chinese company trained a text to video concept using shutterstock footage, I still remember using it and being wow'd)




No one, and I mean absolutely no one from the AI community thought we were going to see Sora's quality for another 1-2 years. And OpenAI, which did not have a text to video model out at all, pulls this out of their ass.