A model of Misa Campo, one of my long-time favorite models.
You should use this LoRA with the epicrealism model, using this prompt structure for the positive prompt:
Misa Campo, a sexy Asian woman, wearing <something> <lora:misacampo:1>
This generally gives good results, along with the badhandv4, OverallDetail, BadDream, and UnrealisticDream negative embeddings. Below about 0.9, the face starts getting mangled, the anatomy goes wild, or both.
Description
Pruned the training dataset to make the face more consistent.
FAQ
Comments (3)
WHY do you give instructions what model and negative embeddings to use? The whole point of a LORA is that it works no matter what model, and I can tell you this also works in deliberate and cyberrealistic. Not very clever.
It is simply a suggestion as to what the LoRA is trained against and ways to help its shortcomings. A succinct form of documentation of its optimal use, if you will. Want to use another model? Go for it. Just don't expect the absolute best results. I've personally found that LoRAs make good images more often on the models they were trained against. Hence, the documented suggestion.
@83catsonthemoon OK point taken I can see where you're coming from. This, however, works good in other checkpoints!