This LoRA is meant to re-create a "laced" or "lace up" jockstrap. It is inspired by this JOCKMAIL jockstrap, but the dataset contains a variety of similar jockstraps. Thanks @blackfrostjr for the idea.
Usage
The LoRA can handle sub-concepts to varying degrees of success. You can try out the below tags to customize the jockstrap.
Use
<color1> and <color2> laced_jockstrapto mix and match colors.Use
<color> laced_jockstrap, <color> lacesto differentiate the jockstrap color from the color of the laces.Add
<color> striped laced_jockstrapto add stripes to the waistband.When generating the jockstrap from the back it will sometimes add a cross-lace to the waistband. This wasn't intended, but I thought it looked cool. Try just using
jockstrapinstead oflaced_jockstrapif you don't want the cross-lace.You can try
holding lacesand/orundressingto generate a person holding the laces, but this isn't as strongly captured.Add
untiedto generate the jockstrap with the laces untied. It can get a bit messy so multiple generations might be needed.
Description
Use a higher weight if the LoRA generates other clothes like briefs and shorts instead of the jockstrap. Mentioning the subject's butt and nakedness is sometimes needed in order for the jockstrap to properly generate from behind or the side.
Trained on Flux.2 Klein 9B Base with Prodigy and a learning rate of 1 with AI Toolkit by Ostris.