SD 1.5 VERSION: https://civarchive.com/models/274183/facesitting-sd15-girl-sitting-on-face
v1.1
Optimized the training parameters to make the model much smaller in size with the same dataset.
*All of the preview pictures for v1.1 can be imported into ComfyUI to see exactly how they were generated. No extra embeddings, upscalers or detailers! Just a hi-res fix to get to a reasonable resolution.
I highly recommend extra embeddings, upscalers, and detailers to get the most out of your composition!
A LoRA trained on high quality images to produce satisfactory facesitting images in SDXL. Multiple angles supported, but results may vary depending on the base model and strength.
Clip Skip: 2
Optimal Strength: 0.5-0.8
Trigger word: facesitting
Add other terms as well such as "sitting on chair" or "kneeling on a bed".
The LoRA attempts to take over any sitting, kneeling, or squatting actions. Adding the term "boy lying on floor" or "boy sitting on chair" may help if the male is not appearing. If the male is "sitting" against the object, the LoRA should re-arrange the scene to make his face the seat.
Description
Optimized the training parameters. Smaller file size with the same dataset.
FAQ
Comments (15)
LOOKS BETTER MORE. NICE WORK.
CAUSE I KNOW It is not easy for AI to identify the position of a human's face when lying on their back.
GO ON DUDE!
Thank you for the support!
Thank you for this Lora. I try to make 2 girls, one lying and one sitting but it almost each time creates a male on the floor. I cannot force AI to create female`s head with female`s hairs. I made over 1k images with maybe 10 ugly girls total, lol.
Nice find! The model was trained on a set of strictly women sitting on men so that may be why the hair is always short for the one on the floor. Matrix prompting may help (where you can specify different prompts for different areas of the image).
I'll see if I can find some high quality images where the girl is being sat on by another woman for the next dataset!
I'm playing around with this right now, and I'm enjoying it! Top Notch!
Thank you very much! :)
How can I stop it from always doing naked ones? Im trying to keep clothed
Put naked in the negative, something like naked:1.4, and in the positive specify the type of pants “wearing black yoga pants”. I do have images with blue jeans, suit pants, and dresses in the dataset, so it should work with enough interations!
I add that directly into the normal prompt area? or do i have to use like the xyz prompt thing, im new to all this lol
@theman7117 In the normal positive prompt box :). I also tagged all bottomless pictures with the tag "bottomless", so putting that in the negative may help as well. The LoRA is a bit biased towards naked bottoms as most images in the dataset contained that.
This was my first LoRA I had satisfactory results with so I am new to this as well!
Could you add this to tensor art please? Would really appreciate it
what are the checkpoints commonly used with this?
top notch!
I cant seem to get it with the man's hands restrained, they are always grabbing her legs. Any tips?