TLDR: a year ago AI video was garbage. Today it’s almost as good as one that would cost a few hundred thousand dollars to pay a human production team to make (according to someone who’s professional work is creating those videos).

It’s not quite there - hands glitch out occasionally. Sometimes animation doesn’t quite line up right (e.g. walking might skip a step) but it’s 99% there and and the improvements over the last 12 months are astounding. That last 1% surely won’t take long to close.

There was a landscape drone video from a helicopter that looked absolutely real.

Note this is not publicly available yet - OpenAI said they are still working on safety features to reduce the risk of it being used to create content that they want no part in.

I’ve asked Gemini for a summary and it’s pretty spot on:

This video is about AI generated videos and how they have become very realistic.

The speaker, Marques Brownlee, discusses a new AI model called Sora that can generate videos from text input. He shows examples of videos generated by Sora, including one of a woman walking down a Tokyo street, a car driving up a mountain road, and a litter of puppies playing in the snow. He points out that these videos are still not perfect, but they are much better than what was possible just a year ago.

He discusses the implications of this technology, both good and bad. On the one hand, it could be used to create fake videos that could be used to deceive people. On the other hand, it could be used to create stock footage that is more affordable and accessible than ever before. Brownlee concludes by saying that this technology is still in its early stages, but it has the potential to change the world in many ways.