Hairpulling Doggystyle!
Doggystyle, but with hair pulling!
All videos were tagged with " dgystylsx. A naked woman and naked man performing sex. The woman is on all fours and looking up. The man is kneeling behind her and shaking his hips. The man is holding her hair tightly. His penis is moving inside her vagina. side view. "
I worked on this for a while now and had technical problems with greyed outputs and strange checker pattern over the whole video. Then I dropped the training resolution (and rescaled videos) from 640x360 to 228x128 and suddenly everything looked good =) I suspect my mid quality training videos were too crappy for 640x360 and I trained in alot of compression artefacts instead of the actuall content.
Description
FAQ
Comments (22)
What is the size of your computer's memory and graphics card
I have an RTX 4090 with 24GB VRAM and 64 GB System RAM. But this training only used around 12-16 GB VRAM because of the small traninig resolution.
The lora is amazing, works extremely well with character loras!
good to hear =) Haven't tried character loras with it myself so far
I think I am seeing that checker pattern come up sometimes... never seen that before... otherwise, good work!
hm... never saw it on this Lora. I'm using the standart comfyUI workflow and lora strength 1 without any other lora for the examples. What are your settings?
@WhatTheGuy The workflow I am using does a LatentImage upscale... it can mess up sometimes creating wavy-ish lines around object... this time though there were square dots all over the place, pretty odd. Alas i think it only happened once so maybe nothing to worry about.
Great Lora, thank you for sharing!
If you don't pull my hair it's not any good 🥰 Thank you for sharing!
Haha, yes! Pulling hair was always missing in the other doggy Loras. I had to fix it ;)
Great lora, is there any trick to get them closer to the camera? When I describe the background, the women will be far away , causing the face to be blurred and deformed.
I usually start with 'closeup' then. If if doesn't work you can give it more weight with (closeup:2) for example.
close shot or (close shot:1.5)
There's something about training at low res that seems to be better for action loras. dtwr's excellent loras all use 244pt.
yes, it was his guide that made me try out a small resoltion!
Great lora. What tool is used for training? I have played around with different tools and almost nothing works well.
Thanks =D I used Musubi Tuner! Small video and training resolutiton for motion Loras did it for me! Character / Objects Loras worked well with high resolution of 1024x1024
Thank you for your reply. I have already tried diffusion-pipe and OneTrainer and both didn't work well for me locally. Misubi was the last in my check list
@AstroWeasel Musubi rocks ❤️
Curious what strengths and sampler you guys are using for this
the standart built in comfyUI workflow without changing any settings ;)
Hi god, isn't there possibility for a POV version ..?