Thank you for your feedback. I've found that training LoRAs using cunnyfunky's works often leads to various strange issues, such as the character's head disappearing or not being within the frame, and distorted limbs. This phenomenon seems to be common in cunnyfunky-style models, whether it's a LoRA or a fine-tuned checkpoint. This time, I tried various approaches, including cleaning the dataset, changing captioning methods, and adjusting training parameters, but in the end, the results were either worse or showed no significant difference. I currently have no ideas on how to resolve these issues. The version I'm uploading now is one that performs relatively well among them.