Yeah. I just read a bit about it and looked at some examples, and it seems CFG>1 is a hack to get sharper images from a given model, at the cost of giving them the "AI look". As models get better, we'll be able to use CFG=1 and much of the problem will go away.