As usual, I was surfing the internet and saw 4 pictures. I thought I needed to do something, so I cropped these 4 pictures into 20 images for training. I could only train these small datasets on my 4070, but it also took me two hours. Maybe I will use the images generated by Lora to train the second version later, but who knows, I think it's very likely that there won't be a second version left. I'm still satisfied with the performance of LoraUsing ADetailer

Files with SHA256: a3f363a0c52856f0bbaf0634d9dc7dc08121592f75d1d33b8e7d40631f1b48a7