This LORA was a real struggle to train due to its unique horns, and perhaps because the rest of the design was too simple, it kept running into all sorts of problems every time it generated something.
In transitioning from the TEST version to V1, we significantly increased the number of profile training images. This has made it considerably easier to generate side-view images, resulting in a LORA model that is now sufficiently practical for real-world use.
Include these prompts at a minimum :
white eyes, black sclera, no pupils, blue skin, bald, horns, teeth, pointy ears, muscular, veins,
clothes(?):
weapon, blue club, club(weapon), spike club, blue aura,
Negative:
extra arms, multiple arms, underwear, tail, dual wielding,
Without using negative prompts, there is a tendency for the model to become multi-armed.
Additionally, due to emphasizing “horns” during training tagging, it is susceptible to Furry-related prompts and models.