With LLMs creating plausible images and even videos so quickly, I wonder in how far they replace ray tracing and if their lack of actual rendering will keep giving away the fake nature of the results. We all know to count fingers but how well do they handle reflective surfaces? Are rooms with a mirror or two fake-resistant?
Discussion
I’d say it’s already there.
Have a look at the Two Minute Papers YouTube channel where a lightning specialist talks about AI:
https://youtube.com/playlist?list=PLujxSBD-JXgk1hb8lyu6sTYsLL39r_3bG&feature=shared
There is no mention of AI or LLMs in that video. I know you can create renderings from 3D models but that's not how LLMs do it. Thus my question about tricky light mappings as these are generally not done acurately using LLMs.
Sorry, I meant to link to the whole channel and I just saw that it degenerated to an nvidia martketing platform.
The kinds of video I meant is this: https://youtu.be/gnctSz2ofU4
This was already six years ago and there are many like this in his channel.