Pussyjob, a.k.a. pussy grinding, etc. Trained with close-up and perspective videos.
Usage
Workflow is included in the still images in the example
Trigger words:
Pussy job, rubbing pussy on penis.pussy rubbing against the man's wet erect penisRecommended parameters
LoRA strength: 0.7 - 0.8
Guidance scale: 9
Notes
It tends to make her body wet or oily
Also it tends to grow pubic hair
Training settings are not available for now since I am still experimenting
Description
Initial version
FAQ
Comments (23)
can you share your workflow to use lora? i cant connect nodes
You can download the png image in the example (scroll to the right). Just drag and drop the file on ComfyUI. The workflow is embedded in the image.
Nice Lora! But I'm getting blocky images and glitched frames sometimes. I'm using the native Comfy nodes for Hunyuan. First Lora that gives me those problems.
I noticed a similar issue, especially in a first few frames.
It has to be something to do with the dataset. I will fix it in the next version. Thanks for reporting!
@2929dance hah, I just opened the videos in an external player and everything is fine! The problem only happens inside the comfyUI player. Now it looks perfectyl fine!
This is awesome! Please make more video-loras!
So, this is my first time trying to get a video workflow going and the workflow from your sample images stops at some point and asks me to install triton. I have no clue where i have to install it and since im a total noob, i simply tried with "git clone https://github.com/triton-lang/triton.git" in my models directory of comfy, but that did not work. Can anyone help me out?
It is not required, but it makes the generation much faster.
As far as I remember, I followed the instruction here to install triton and sageattention.
https://www.reddit.com/r/StableDiffusion/comments/1h7hunp/how_to_run_hunyuanvideo_on_a_single_24gb_vram_card/
Thanks, that one did help in setting everything up! But i think i might have to try again with a complete reinstall of Comfy or something. Whenever the process gets to the HunyuanVideo Sampler node, it starts to take ages to get anything done. So far it only got to about 28% in more than 30min while GPU and VRAM are at constant 100% usage, even though im using a 4080 super. So my guess is, that i still messed something up or that something about my system is not working properly :/
amazing work, however im new to comfy ui, is there any lora or other add-ons that is recommended? and anywork flow ?
Just wanted to say thanks for this. It's easily my favorite NSFW HunYuan LoRA by far. Awesome job.
Did you have any vae problems? i can't get passed it.
@finalfight - I'm probably of no help. I let SwarmUI choose... and I have both the fp32 and the bf16 VAEs in the folder. I don't even know which one it uses. Furthermore, I haven't used the lora since I swapped over from using the fast model to using the full bf16 model. The fast model is pretty horrible, and my outputs from most loras were pretty bad before then.
I don't know what the lora looks like with good settings because I haven't tried it out again, and I'm busy training loras all the time. Sorry that's not very helpful. Hope you get it sorted! Troubleshooting these things is one of my least favorite things in the world.
But thanks for the reminder. I need to test this out because I REALLY like it.
@Gongoloid Ah dang, thanks anyway
How is the LORA, with diffusion pipe or other? formed, what are the necessary PC capacities?
Which vae file are you using? Stops at "HunyuanVideo vae Loa"
If it gets stuck at loading, it's highly likely that you don't have enough VRAM. Try finding a workflow that supports low VRAM.
@2929dance Sorry, pretty new to Hunyuan, 4090 is not enough?
@finalfight hmm, then it should be something else. I'm also using 4090
@2929dance Damn....thanks anyway, that's good to know your using a 4090 at least so I know its not OOM.
I like the back and forth movement. I can't find a cowboy lora that has such good movement ... hint, hint :)
Dude, thank you. Over time this has become one of my top 5 HY loras when I want to get spicy. Love it so much.
But the Wan analogue is weak sauce.
It is just ... bizarre and not realistic enough to be fun.
So my question for you is "are you gonna train a pussyjob for Wan?"
and if not, will you upload your training data, and I will do it and upload to civit?
The lora that got me into using video models
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.