Vers issues:
V1T0 Maybe he draws the mouths crookedly, I think...
Additional train info:
Trained on: Illustrious-XL
Based on deep edited and enhanced Aoi Ogata arts dataset.
Description
Damn, in the previous version I accidentally used a min_timestep 700, which caused the LoRA to train more on large details than on small ones. The poses were too deeply ingrained in the LoRA.
The config was incorrect for this LoRA.