Maybe I shouldn't try to train multiple features in a single LoRA. Should I really make three seperate LoRA for this?
Making a lora with one feature means that I have to connect multiple LoRAs later when I generate the image, and it tends to result in hiccups.