Hacker News Reader

Generating a video offline and doing it in real-time are very different problems

You can spend days generating one frame of the video offline

Whereas realtime you need to generate a frame in the magnitude of 30-40ms

I think it will come down to whether the calculation to render a scene in a game engine is most costly than rendering one through an AI model, right?

I'm confident that rendering in-engine, in realtime, will always be less costly than calling a remote API. You also get the benefit of precise control over scene and content, rather than the generic results of an LLM.

It's not as if the high end of modern generations aren't already capable of photorealism - or has everyone already forgotten the GTA 6 trailer?

The only reason to choose AI is to be able to fire artists, which is what the calculus will actually be. Not whether AI is a superior solution (it isn't) but simply adequate enough that studios can made up for the drop in quality by cutting employment and still make a profit.

Reply on news.ycombinator.com