The Lora greatly improves lighting and character aesthetics ,use example pictures of my own workflow to train.
sdxl version
Recommended Settings:
Sampling Method: DPMPP_2M
Schedule Type: karras
CFG Scale: 5-6
Recommendation weight is between 0.8 -0.95
Anima version
Recommended Settings:
Sampling Method: er_sde
Schedule Type: simple/beta
CFG Scale: 5-6
Description
A LoRA trained based on natural language labeling, and it uses Chenkin's trainer.