Lora is based on several styles: palette knife painting (70% of dataset), oil painting (20%) and adding some style of my 1.5 MaxRos model

Tested with DPM++ 2M Karras (better to use at least 30 steps)