i really think it's going to take much longer than people think for this technology to go from 'pretty good' to actually being able to meet a production standard of quality with little to no human involvement. at this point, cleaning up after an ai is still probably more labor intensive than simply using the cheatcodes that already exist for quick and cheap realism. i expect in the midterm, diffusion models will largely exist in the same space as game engines like unity and unreal where it's relatively easy for an illiterate like me to stay within the rails and throw a bunch of premade assets together but getting beyond NINTENDO HIRE THIS MAN! and the stock 'look' of the engine still takes a great deal of expertise. >https://www.youtube.com/watch?v=C1Y_d_Lhp60