I wonder how much RLHF or other human tweaking of the models contributes to this sort of overstauration / excess contrast in the first place. The average consum

I wonder how much RLHF or other human tweaking of the models contributes to this sort of overstauration / excess contrast in the first place. The average consumer seems to prefer such features when comparing images/video, and use it as a heuristic for quality. And there have been some text-to-image comparisons of older gen models to newer gen, purporting that the older, more hands-off models didn't skew towards kitschy and overblown output the way newer ones do.