*Foams from the mouth* I never seem to learn... By now i should have learnt that i am mostly shit on concepts. Worst of all after I finished I realized there's another similar LORA that also works ok-ish. This shit also seems quite compatible with that one.

So here i am one week latter with a concept LOCON i am barely ok with. I get the feeling that if I had received this lora at the very beginning i would be happy with it. But after all the effort put on it, it just feels that the damn holes in it just get bigger each time i look at it.

Anyway... I will add two galleries a Pure one and a mixed one using my Loras.

Unmixed it seems to ok until .9ish, recommended setting:

• <lora:kissing_penisV8:.7>, <one of the poses below>

Mixed it seems ok until .7ish, recommended setting:

• <lora:kissing_penisV8:.5>, <one of the poses below>

I split the concept into 3:

• kissing_penis_pov: Just the penis in pov mode seen from the front. Seems to work fine from 512x512 to 512x768,. You can make the image broader to a point.

• kissing_penis_from_side: Profile view sometimes includes testicles. This sometimes needs to be a bit wider so probably 512x512 to 768x768

• kissing_penis_paizuri: Penis coming out between her breasts. This one needs to be longer to accommodate her breasts so 512x768 to 512x1024

Forgot to mention it but spoken_heart was also explicitly trained cause it appeared a lot..

I added some reg images to decrease contamination a bit. It kind of works? You shouldn't reach heavy style contamination in the lora mixes until you hit .7 weight.

Of course as always with concepts is a 3 way battle between reproducibility, style contamination and burning.

As Always I will add the dataset and tag summary for people want to experiment and as always i hope to forget i made this at least for a couple of months.

Leave a comment if someone is interested in a normal LORA version instead of a LOCON, training it would be trivial and will give slightly different results if nothing else.