Presenting DR34MSC4PE's second Hunyuan LoRA model: Blowjobs!
We hope you enjoy!
===========================================================================
Recommended Settings
These resolutions seem to work pretty well:
640x480
960x720
720X544
544x960
Steps: 20-40
CFG: 4-10
dpm++2m/normal or beta
euler/normal or simple (euler_a also works)
ipndm/beta
deis/beta
Experiment with your own settings to see what works.
Sample Prompts:
bl0wj0b, pov shot of an asian woman with fair skin and straight shoulder-length dark disheveled hair with blonde streaks. She is nude as she performs a blowjob on a man's erect penis. She strokes the shaft of the penis with her right hand. The scene takes place inside the cockpit of a spaceship, illuminated by led lights from the control panel. The camera angle is a pov shot focusing on womans actions.
bl0wj0b, A pov shot of a young russian woman with fair skin and straight shoulder-length dark disheveled hair with blonde streaks hair. She is nude as she performs a blowjob on an erect penis. Her perky breasts are exposed as She strokes the shaft of the penis with her right hand as she takes the penis into her open mouth and sucks on it. The scene takes place inside a jeep wrangler as it travels through the wilderness, trees can be seen out the window as they pass by, The camera angle is a side shot focusing on woman's actions within the vehicle. Cinematic, ARRI Alexa, realistic, masterpiece,
Check the attached dataset text files to see more examples.
===========================================================================
DR34MSC4PE is
c0ur4ge (trainer/qa/inference code) /
ERA5ER (captions/tooling code/data/qa)
Be sure to check out our FLUX model - C4PACITOR!
Like our work? Buy us a coffee: https://ko-fi.com/dr34msc4pe
Please generate responsibly!
Description
Trained on BF16 with about 60 images and 150 short video clips. See attached zip for sample captions and workflow.
FAQ
Comments (19)
We use the same file-save-prefix format, we must be some cool dudes
Also this is an amazing lora, I bumped up the resolution and everything to appreciate it. The little details, it brings through some personality the others haven't so far.
Thank you! Yes, pumping up the resolution can get pretty great results. Just takes forever to generate!!
I'm probably missing something in the setting, or my workflow kinda don't work with this lora, anyway zero good result on my side, like not even close, your example vid looks quite good though.
Just uploaded a few more samples with included PNG's, as well as included workflow I've been using in the attached Zip file.
Admittedly, not every generation will be a winner, but hopefully the workflow helps a bit.
@ERA5ER i'll give it a try thx
As with most of these loras, I've had to use the creators prompting almost exactly to get the right results. In the case of this lora, cheating on the low resolutions resulted in nonsense. However in the resolutions recommended, it works perfectly. Ive been using the prompt within the OP's flows embedded in the photo. I don't think it's that much different
Awesome the 50th blowjob lora. Why no one has done even one footjob lora?
Provide the dataset and I'll train it.
It takes a lot of effort. You could just pull off any "feetjob" videos from the internet, but you're going to want higher quality, high enough variety, and having it all split out into photos and then properly tagged... You know you should probably read all the training tags, hell probably have to rewrite them... there goes the afternoon lol
the thing is that every lora has a different feeling to it. Some feel more like a everyday amateur clip, others like high quality studio productions. Some are more rigid in their angles, others have a bigger variety of angles. Of course entirely new concept are more exiting, but having a choice of the same concept is valuable too ;)
Or just a general "using hands on penis" lora. The only handjob lora is super vigorous and moves too fast and is too repetitive.
@ERA5ER I tried making my own but havn't been able to get it to work, Im afraid Im not the best one to come up with a dataset. That being said I have a lot of videos but it's not easy to share. I just wish someone who is passionate enough to make it would do that, yes it's difficult if I knew more I would keep trying but I've given up for now. Im a bit frustrated no one has put one up yet for Hunyuan. I offered to pay before and still no one lol
@google910 It's still early, Hunyuan barely came out a month ago. Give it time. I expect feet are as tricky as they are on FLUX.
Foot job is here https://civitai.com/models/1178240/hunyuanvideo-footjob
@ERA5ER fr, patience is key or make it yourself : )
Not as hot
Very good! it doesn't really work well with the Hunyuan Intermediate V2V workflow i use so i tried a different one. But i got results at 512x512 at 97 frames and 720x512 at 73 frames. will post some results soon.
did you ever post your results from V2V?
Yes I have gotten very good results I have posted a couple of results over the past week.