This is a model I created that was trained on a curated dataset of facials. I highly recommend using juggernautXL for the main model, and then using this lora at somewhere between (.4 and .6). If you use it with the base SDXL model you will need to increase the strength as it really tries to keep things sfw. Tags like "kneeling", "facial", "tongue out", "cum on face", and "perfect facial" tend to get the best results.
If you have any questions leave a message.
Description
First version. Doubt I will create another one unless there is significant push for it as this one functions very well.
FAQ
Comments (12)
O M G ! yes! thank you brother! god bless you
Work good?
@Chimaeraic_Generation holy mother
Please tell me there's plans for 1.5 as well? Because hot damn, great work!
I could train a 1.5 model as well using the same dataset, don't have the buzz to do it currently though.
Sidenote- You can use this model at low strength (around .3 or so) on a generation of a person at night to get the subject lit up as if there's camera flash. Interesting side affect of the type of images used to train that can actually come I'm handy in sfw generation.
Thank you. Any trigger words for cum on cloth/ clothing?
The images got automatically tagged by a tagger. Can't remember if there were any specific to clothing but I would imagine with a bit of nudging you could get it there. Although any cum that's not directly on the face is not what it was trained for specifically.
really hard to get cum on breasts without cum on face. you have to turn up the strength too high and it wrecks the rest of the image. using GFPGAN face restoration helps a little (it clears faces) but it still doesn't help the fact that you can't get much cum on the chest.
wow it kind of felt weird typing that ngl
I might be able to make another lora specifically for cum on breasts, but this one was trained specifically for the face, so this kind of makes sense.
Can't make it work with a face lora they seem to outweigh each other, any help?
I suggest trying https://github.com/cheald/sd-webui-loractl
then you can do something like
<someLora:0@0, [email protected]>
which would someLora applied with 0 weight at start and only start applying it at 0.8 steps (80% finished steps generation)
so you can get your face lora first - then apply this one towards the end.