I couldn't find a similar lora for this position except the "sword swallowing position" but that didn't have the effect I was looking for.
Trained on 170~ images, generated captions with each edited by hand. Works best, in my opinion, at 0.6-0.8 weight for getting the position, but it also carries some lighting effects that I really like at the lower 0.2-0.4 weights.
V2:
Trained on 125~ images, generated captions with each edited by hand. Works best, in my opinion, at 0.6-0.8 weight for getting the position, but it also carries some lighting effects that I really like at the lower 0.2-0.4 weights.
Removed all images from training set that were not under boy, facing girl. Replaced those with the trigger word "blowjob from below"
I took the model down because I was seeing some major flaws in the generations after even more testing. I think that because the face is always obscured by the penis in this position, SD has a hard time keeping the facial structure correct. The refined training selection seems to have helped in that the face will mostly be symmetrical on each side.
I had more success when I used the good generations as canny edge maps and then applied those maps for future generations with controlnet.
Attempted to differentiate positional arrangements with the following trigger words:
"under girl, facing boy" : viewer is under the boy, girl is facing viewer (images tagged this way, but I could not get it to work with my checkpoints based on women as the main focus, so your results may vary)
"under boy, facing girl": viewer is under the girl, boy is facing viewer
"behind guy, laying on her back": viewer is behind the boy, girl is laying underneath. May work better with "69" as additional trigger.
Blowjob depth trained with following variations:
"just the tip" : mouth at the tip of the penis or just around the head/glans
"blowjob" : tip inserted fully in mouth
"deepthroat" : halfway+ penis in mouth
"to the hilt" : entirely swallowed/all the way to the base
I got varying results and the weights may need adjustment.
Additional positioning trained through the following:
"hand on another's head" : one hand on girl's head
"hands on another's head, grabbing head" : both hands on girl's head (interchangeable)
"close-up" : close up shot with face and penis
Detail adjustment:
"lips around penis" : if lips were pronounced in initial image
"dark-skinned male, dark-skinned penis, interracial" : typical traits to adjust male skin color
Description
Removed all images from training set that were not under boy, facing girl. Replaced those with the trigger word "blowjob from below"
FAQ
Comments (1)
Seems to work well enough, although I’m still figuring out how to get it to do ‘to the hilt’ deepthroat images. I would love to see something like this in an XL model, particularly Pony Diffusion XL v6