like when you "train" an ai, you aren't literally starting from scratch, you're just introducing a bias for what you want the algorithm to generate lmao

Id heard of like low power usages that just recreate yr dataset (memory serves it was an in-betweening thing used on Spider-verse but that may be closer to "machine learning") but like, I don't know if those are mass adopted yet