CockUnderXL

CockUnderXL (alpha01)

This is my first publicly posted Lora. This was a way for me to learn how to make a SDXL Lora, and I ended up testing and tweaking it to the point where it works.... sort of... most of the time.

This is a concept Lora that produces a cock under clothing for SDXL models, generally those that are photorealistic and have some idea of men and clothing. It generally reliably generates a "pants snake" - an erection that lay along the body.

The main trigger words are:

"cockunderxl" - main trigger for all possible variants that follow below. This trigger was the first word in all image prompts.

"pantssnakexl", "pantssnakexl-R", "pantssnakexl-L" - reliable triggers for pants snake. pantssnake-R and -L (pointing to the right or left) were attempts at reducing "double penises", two penises on the same body and pointing in opposite directions. In my testing, using a dynamic prompt {(pantssnakexl-R:1)|(pantssnakexl-L:1)} was somewhat helpful. "pantssnakexl-D" and "pantssnakexl-U" (down and up) also exist but are not as reliable.

"tentingxl" - a less reliable trigger for a tenting erection. More successful when the subject is not standing. "tentingxl-R", "tentingxl-L" "tentingxl-D" and "tentingxl-U" exists but were not very reliable

"wettentingxl" - is a trigger for a erection under wet clothing, not reliable.
"hardcockoutofwaistbandxl" - is a trigger for an erection still covered by underwear that appears out of a waistband, poorly represented in the data set and isn't reliable, but can appear in some gens.

In addition to unique tags, the natural language phrases "erection under clothing" and "erection under clothing" were used. In context appropriate situations, the phrase "translucent underwear" was used.

"clearvplxl" - a situation where it was less obvious if the penis was erect, and more of a VPL. Distinct underlying shape under clothing however, was clear.

"drapedvplxl" - under represented situation where cloth or clothing was draped over a cock which was less definitively erect.

"wetvplxl" - underrepresented situation where a vpl was visible over wet clothes.

In testing VPL tags tend to still produce erections. natural language phrases "penis outline" "VPL", and "dickprint" were used.

My recommendations are to use with a realistic checkpoint that can generate men with underwear. The strength of the lora and the reliable trigger words depends on the checkpoint and the checkpoint settings, but starting out at 1 is generally a good place, but adjustments will likely be needed.

The model was trained with a mix of unique tags first, then ViT-g-14-laion2B-s34B-b88K natural language phrases, and then finally ViT single word phrases.

This model is trained almost entirely on pale skinny twinks, so it is the extreme opposite of diverse. You can try generating other types, but it might be highly resistant.

Using Adetailer is highly recommended. Double cocks is still a problem.

If you incorporate this Lora into your mixes, I would much appreciate a shoutout or credit. I do this for fun and education, please do not sell or use this lora commercially.