Lora works overall pretty well but it tends to randomly pick any hole and hand variation, instead of what's prompted, maybe a more specific tagging could help.

I've heard of finger-puppets, but...yeesh!
\:o

Can you share the data please?

Although it's for PonyXL, it works quite well with IllustriousXL-based models. (It seems to work better with a higher weight, though.)"

Appeared to be exactly what i was looking for, turned out not to be. It's badly limited by the dataset (which is based on what I assume is your preferences). It's great if people's tastes match yours perfectly, but harms useability otherwise.

With no additional lora, it -strongly- pushes middle-aged, a slightly chubby body, reclining position, bed/couch, and probably some other stuff.

The vaginal and anal tags also don't appear to make any difference, as far as I can tell it just picks randomly (i.e. 'vaginal' in positive and 'anal' in neg will still give you a fairly even mix).

Some of that could probably be overpowered by other lora, but didn't care to spend the time tweaking weights for testing.

Briefly tested with base Pony 6, Pony realism 22, and juggernautXL 11