Lora version of my Impregnation/Tanetsuke V2 LOCON as some people simply prefer simple LORAs. I don't really see much difference but the LOCON seems to work a bit better with some misbehaving models likely as they export more layers.

Anyway this started on an ill conceived(pun intended) idea a while ago as the other similar loras were extremely style heavy, limited in working positions and mostly overriding hairstyles as they were created with very limited datasets. So I embarked in the route of creating an all synthetic dataset as varied as possible. It was a pain in the ass and resulted in my inpainting LORAs. Using that I inpainted The dataset for the V1 LOCON. At that point i was a bit burn out and said good enough and tried to forget it even existed.

At some point I started retraining my first LORAs and as expected i reached the point in which i once again stumbled on this... So this time I gave a manual cleaning to the synthetic dataset. The main problem with this concept is that SD and NAI have absofuckinglutely no idea of what it is and any contribution of the model is for the worse. So I had to basically sear the concepts into the lora with sheer spite. I had actually forgotten about it.

Anyway, This seems to have less style contamination than the V1 LOCON, but the LORA version seems to be a bit more flaky that the V2 LOCON so you might have to push the weight of the lora above 1.

I might revisit this crap for a another try in 6 months when i have forgotten how annoying it is.

Seems to work ok with <lora:TanetsukeV10:1>, (Tanetsuke),

In the dataset i included the following poses which i hope will be correctly displayed with their bubbles:

• cowgirl facing viewer

• reverse cowgirl facing away from viewer

• standing leg up facing viewer

• doggy from behind facing away from viewer

• Some prone bone position

• mating press/missionary

• A couple after sex poses

• Standing no sex pose

I'll add the dataset so you can check it and get an idea of what should be posible.