Example prompt:
missionary sex, POV overhead view of a woman lying on her back with her legs spread having sex with a man. She's on the couch. She is moaning with pleasure. A man is thrusting his penis back and forth inside her pussy rapidly at the bottom of the screen. The camera is zoomed out and holding steady. Her auburn hair is long and curly.
Version 1.2 Updates:
I used the same training videos as before, but this time I blurred the faces. I hope this created a non face altering version, and it definitely does seem to work better with character loras, and the facial variety also seems a bit better.
I used https://github.com/ORB-HD/deface to blur the faces in the training videos, and then added "woman with a censored and blurred face" to the captions.
Version 1.1 Updates:
I added a few new videos to train on, zoomed in more on the movement itself to hopefully train that a bit better. I also lowered the learning rate to 5e-5 and bumped up the number of repeats to 30.
The result seems to work much better at lower strengths and hopefully better with character loras now.
Note that this training took 8 hours compared to v1.0's 1.5 hours. There's probably some sweet spot for learning rate and training time to get good results, but it would take more experimentation to figure it out.
I don't feel comfortable sharing the exact mp4s I trained on, as they were just ripped from online sites and I don't really have the rights to distribute them. However, I will include my training data for the config files and captions so that other people can more easily get into training. I was surprised at how quick and easy it was.
I included an example workflow in the training data download (I can't find a better way to upload the workflow), which shows how to make it work nicely with multiple LoRAs, and has dynamic prompt support.
v 1.1:
I trained on a 3090 using 11 3 second videos (24 FPS, at least 50 frames each) and it took around 8 hours to do 20 epochs with 30 repeats.
v 1.0:
I trained on a 3090 using 8 3 second videos (24 FPS, at least 50 frames each) and it took around an hour and a half to do 20 epochs with 10 repeats.
Description
Non face altering, so it has better facial variety and works much better with character loras.
FAQ
Comments (26)
1.2 is next level
You said you are using small video, what kind of txt file prompts you are using for each video? i am trying to make one for side doggystyle but cant make it work. Any advice would be great, as i havent buiilt lora for videos only for images.
nice update, all your loras work well, thanks, we need some good pussy detail/shape/action loras if u feel inclined ;)
Any idea why my outputs aren't the same? I'm using the same model/vae/clip/llm, same prompt from workflow, your lora, same settings as in the worklow, but my outputs look different (different lighting, more errors like extra hands, slightly less detailed).
Honestly, I have no idea. I just loaded up the included workflow and generated without changing anything and I get something very similar to the sample here (I didn't save the exact seed I used from before, so it's not identical). I don't know how much of an impact different hardware makes, but I hope differences there wouldn't cause the type of changes you're talking about.
Though occasionally, it does put the woman's hands both by her side and grabbing her legs, so it's not perfect with that stuff for me either, though it doesn't do it very often.
why do their faces always look like they're being tickled? What's so funny about sex? Aren't their faces supposed to reflect the emotion of pleasure?
What do you mean? You can prompt for whatever facial expression you want. You could probably make a lora if Hunyuan doesn't understand exactly what you're looking for.
You never made a girl giggle into a nut?
wow, just found out this exists! can it do image to video? what image size? 1024x1024? or 720p? good job man
I don't know anything about the image 2 video stuff that exists for Hunyuan, as I'm mostly just waiting for the official model to come out. Of the fan made stuff, I don't know how well it works with loras and things like that.
Everything I've done has just been text to video.
No idea about specific LoRAs, but even without loras on a 3080 12gb img to video is basically impossibly slow even at lower resolutions. Not worth the time. Maybe on a 5090..
on a 3080 I use 320-480 width and upscale - it's going to be slow but for now it's relatively speedy considering it's running locally.
There is a reason this is the highest rated position lora and it is because it is by far the most consistent with any character loras. And after testing most others, I am confident it is because of the training method you detailed in your description. I think what you did should be common practice for any lora meant to associate character loras with specific actions.
this is crazy
Amazing LoRA! Any chance for a Hunyuan POV Doggystyle?
can we combine this with character lora?? any tips or guides to it?
The training images download should contain an example workflow. It has an example of a character lora that you can enable.
i2v possibe?
dicks are thinner than previous versions. Too thin.
Pls is there a way to set speed of the thrusting? I tried different prompts, but nothing works. All I would need is slow, medium, fast. That is all.
The training data has some things tagged as slowly, and others rapidly. Whether it actually prompts for that well, I'm not sure.
It is weird, but this Lora is the only one that doesn't react at all. It is like it hasn't been added. It really is Hunyuan right ?
Works great for short videos, but there is all sorts of wierdness if used with FramePack Eichi, after about 20 seconds or so, start having things like double and triple hips with vaginas, the person thrusting will sudddenly turn into a body of penis on one side, pussy on the other, etc. Really odd, and am unsure how to fix it other than to just generate shorter videos.
So the distortion only happened on F1 model, using. It works great on the original FramePack model.
I've noticed that sexual acts,the penis only moves slightly in and out against the vagina, remaining outside without full insertion. I'm unsure whether this relates to 'seed' or specific Prompt.😦
Nonsense. This one sticks it farther in than any other LoRA I've seen. I don't even have to prime it with a starting image all the way in.