trained using DRaFT-LV setuppairwise head on joyquality fitted on a dataset of varying cfg generations from the base modelthe preference is cfg 3.0 < 5.0 > 7.0
使用基于JoyQuality的类似DRaFT的后RL方法训练的LoRA,并新增了成对头部
是的,它为所有事物添加了闪烁和粒子效果。