I created 5 versions of this model which all perform fairly well. The first two used different checkpoints which I then used to generate quite a few images. I used those images as regularization images in one model and as additional data set images in the next two. The models that leverage the generated images are considerably more flexible in terms of poses and body types, but require a bit higher weight to get the concept.

I recommend looking at the prompts in the generated images to get an idea of what works well.

Key tags I used in training: tape gag, shockcollar, orange shockcollar, red shockcollar, blue shockcollar, yellow shockcollar, shockcollarremote, rope, close-up, pitlick, shockcollaronly

lora:ShockCollar - checkpoint: homofidelis_v40 - only real images

lora:ShockCollar-Reg - checkpoint: homofidelis_v40 - only real images - regularization of generated images

lora:ShockCollar-Gen - checkpoint: homofidelis_v40 - real + generated images

lora:ShockCollar-CivitAi - checkpoint: SD realistic - only real images

lora:ShockCollar-CivitAi-Gen - checkpoint: SD realistic - real + generated images

Works better with simple prompts. Can produce lots of ugly images with combined with other LORA at high strengths.

The two that probably work best are lora:ShockCollar-Gen and lora:ShockCollar-CivitAi-Gen

Worst is lora:ShockCollar-CivitAi because it is very inflexible, but does produce a grainy photo with beefy guys that is sometimes nice.

This model does not depict a real person and can be used to produce NSFW content on CivitAi.