And now for something completely different. All those face swap loras are boring, I like to challenge myself. I've been unsuccessfully trying to train this lora for SD1.5 for a long time, but when SDXL came out, same dataset produced decent result. Enjoy if you into it.