>>936083886
LoRAs are basically training wheels to guide a gen toward a specific outcome. I don't really use them unless I'm doing something really specific, nor do you really need to if your checkpoint has good adherence and a wide dataset.

As for checkpoints, I like NTRmix, Animagine, Miaomiaoharem_v13 and 5moonIL's pvc figure model for a more 2.5D kinda look. I don't do realism personally, but it's getting better.