Hacker News new | ask | show | jobs
by ahmedfromtunis 391 days ago
It has long been established that Veo has a waaay better understanding of physics, and consistency over multiple frames, than Sora. Not even close.
1 comments

I want to be clear, I don't think Sora looks better. What I am saying is they both look AI generated to a fault, something I would have thought would be not as prominent at this point.

I don't follow the video generation stuff, so the last time I saw AI video it was the initial Sora release, and I just went back to that press release and I still maintain that this does not seem like the type of leap I would have expected.

We see pretty massive upgrades every release between all the major LLM models for code/reasoning, but I was kind of shocked to see that the video output seems stuck in late 2023/early 2024 which was impressive then but a lot less impressive a year out I guess.