This would be my 3rd LoRA I've created. The dataset is very small, and it noticeably struggles to prompt for closed mouths, but is otherwise perfectly functional. (May I suggest the NegPiP extension to help alleviate the mouth issue for now?)

(Trained at 576px with 39 images for ~2100 steps using Prodigy optimizer on terminal SNR FluffyRock e103)

I only trained this LoRA once, but it performed far better than I was expecting. Head shape in particular seems to vary, but can be adjusted with tuft tags. (Mane and beard tags also seem to cause changes?) Arms can disappear if unprompted for, or by using armless tags. Legs can be generated if desired by prompting accordingly. Seems to handle anthro/humanoid body type okay enough. Dataset has little for behind views, so expect that to be imperfect.

Activation keyword is ahitsnatcher; Works fine at a weight of 1.

Recommended positive keywords: toony, purple body, yellow eyes/mouth, legless, ghost tail, fangs (You may want to add ghost and spirit depending on how your model handles those tags.)

Recommended negative keywords: naga, white eyes/mouth, hair (You may need flaming tail depending on how your model handles the LoRA, I've noticed.)