First test of making a Z-Image Base LoRA. It works not too bad, actually, but you will likely have to cherry-pick.
Description
Trained on 28 Images, and captioned using JoyCaption.
Example phrase that might work:
blue underwear branded with "BIG GUY" in white text on the waistband down, revealing a medium-sized, erect penis is long and very thick and visible through the shorts, with the bulge pointing and tenting upwards. Example Workflow w/LTX2:
https://gist.github.com/AnonWaterVole/b35f6370fba06f7f12990dd872958a12