If that's done live, with current generative AI you will get different final model every time, on every machine - that's not a good idea, especially in online games. Again, current AI is mostly generative chaos and way overblown by marketing - it's the easiest to do type of AI and really not very useful for things you suggest, IMHO. And other types of AI are much harder to pull off, which means we are far away from wide retail use of these.
They are plenty fast these days with high amount of cores unless you want to do very advanced things like fluid or smoke simulations - but that's not what players been asking for. Destruction physics had been done and dusted on the CPUs fire many years now (on way slower than modern ones) but lighting was the issue with raster and prebaked lights. With fully dynamic RT it should be easy to do, you don't need GPU for physics at all then, just for lighting it up properly.
Again, generative AI models are fine for creating NPCs with conversations etc. but that is the limit pretty much. And even that isn't as good as manually waiting it by pro writers in most cases, prone to hallucinations and other abnormalities.
It can't be done with current AI tech - maybe in the future, but we are not there yet unless you are happy with really weirdly looking games, each time different and more like horror dreams than reality.
To summarise, I would advise to not listen to marketing rubbish about AI too much - it has its uses but not in a way you seem to wish it to do it.
Last edited:
