Trained with 17 images on Meinamix.
The dataset is not very big, therefore the lora is pretty tight. Weight 0.8 and above is recommended for a nearly screen accurate suit. Controlnet should be used if you want more dynamic poses.
The rubber orange hair piece is very tricky to get right since it's too obscure, so most of the time you'll end up with real hair instead.
Description
Bigger training dataset, slightly more accurate results.