SDXL seems to have somewhat of an idea of what a ballgag is, but doesn't generate anything coherent. This is my attempt at fixing that.

Should work for both photo/realistic and cartoon/2d style images due to a fairly varied training set, can tend to over-size the ball or miss the strap off, and still creates some wonky generations from time to time. You might be able to dictate the color by including it before the tag (ie "red ballgag"), but this seems hit or miss.

Keyword: ballgag
Example Prompt: [photo/cartoon/etc] of [subject] with a ballgag
Possible Neg Prompts: teeth, mask, lipstick, makeup, choker, collar

Trained on dreamshaperXL10_alpha2Xl10, seems to also work well with most other models and Loras.

Note: This is also my first publicly posted Lora after much testing and experimentation. Open to suggestions and knowledge from those who know more.