LoRA produces images of Arthur Pencilgon's outfit from the first season of Shangri-la Frontier, it seems relatively accurate and flexible. I am not sure how to reduce the LoRA size and would be happy if anyone has tips on how to improve the training process, I do it all locally through Kohya-ss.

Face was blurred or painted out in training images, I used adetailer and a character LoRA in most of my sample images except the full body ones where I did not use anything but basic quality prompts and adetailer.

Leave out the prompts for the spear and the hair if you don't want those!