Cock Transformation LoRA for Pony. I just threw the old dataset into the civitai trainer, so it's not optimal. It still has a lot of style bleed-in and anatomy distortion, but it's absolutely usable. I might look into improving this in the future.

Works best with 1.0 strength, but once you get a composition you're happy with, I recommend reducing the strength of the LoRA, and using a low-strength ControlNet to regenerate the image. That way, you can limit the bad influence of the LoRA on style and anatomy without impacting the concept itself too much. Inpainting is also an alternative. Some angles don't work very well, so you might need to inpaint for those anyway.

V0.1 contains three unique tokens, ctf, ctfhead, and ctfglans. Between the three of them, only ctfhead seems to work consistently, so use that.

V0.2 only uses ctf.