My first test of training XL loras using CivitAi. It's a remake of my XiaMo LoRA for SD 1.5.
Due to the nature of the training, text and water marks will often be generated! Sorry :(
Try adding the fallowing to negative prompt to decrease the change of getting text / watermarks :
chinese text, korean text, cardboard box, fake screenshot, poster (object), livestream
A LoRA for XiaMo, a popular chinese girl that makes ASMR