Don't be afraid to strengthen or weaken the tags. Your images won't be destroyed by (tag:2) oftentimes, and you can even juice the loras up to x3 if you want them to fully cut through and overwhelm.
It SHOULD work directly on with standard simv23 lora + F1D_Q_4/etc if bandwidth or hardware is an issue. I need a proper quantified version of Sim3F1D-DD asap, so I'll figure that out on the weekend.
Trained directly with the simulacrum v3 unet and clip_l, likely requires simulacrum_v3 to function, but there's a chance it'll function on Sim v23. Uncertain and untested, trained on v3.
Should handle cosplay, the outfits, the characters, the scenes, the themes, whatever you can come up with. The mixed prompt caption should take care of the rest.
VERY NSFW Sourced 2000 images - ~1700 were deemed viable and used;
Description
Pretty sure it has a ton of women and body types.