I don't think sora is good for anything or anyone other than openai making more money based on hype, but it seems pretty reasonable to me that the company that previously scraped the whole internet (violating everyones copyright) and trained one of the best text models on it, could do the same for video.

I mean, most of the clips they show have some noticable minor (or major) issues, and the "fail" examples are very obviously wrong.

I hadn't even seen the balloon head one, I only watched the clips they published when they announced it.