Deepthroat Blowjob LoRa
Trained on a dataset of 78 clips with 224x224 resolution, 1sec to 3sec
“pov” and “sideview” camera angles (pov seems to be working better now)
22 different source videos.
also i just finished preparing a new dataset with different camera angle, stay tuned for when its ready 😎
* Added v1.2 with "normalized" dataset, i mean all clips split into 3s - 1s clips, and "re-captioned" all, learning rate of 2, also "video_clip_mode=multiple_overlapping" in hunyuan_video.toml this time)
The workflow im using is:
Hunyuan💥AllInOne ▪ Fast ▪ T2V | V2V | I2V* | Upscale ✔️AIO | Advanced Tea☕1.4 from @LatentDream
Examples:
a sideview of a woman giving a deep throat blowjobsideview of a woman kneeling in front of a man deepthroating his penis, she sucks deepthroat style, woman sucks a big cock, she swallows the whole penis, deepthroat all the way, she looks up at viewer, balls deep blowjob, deepthroat, lips on hipsa sideview of a woman swallowing a penis in deepthroat, the woman is kneeling over the penis, she swallows the whole dick touching the testicles with her chin, she sucks it from tip to base, the man grabs her by the hair, side_d3pthr0AtPOV view of a woman kneeling in front of a man deepthroating his penis, she sucks deepthroat style, woman sucks a big cock, she swallows the whole penis, deepthroat all the way, she looks up at viewer, balls deep blowjob, deepthroat, lips on hips, pov_d3pthr0AtHave fun! 😜
***** This LoRA is shared without warranties and with the condition that it is used in a lawful and responsible way. I do not support or take responsibility for illegal, harmful, or harassing uses. By downloading or using it, you accept that you are solely responsible for how it is used. *****
Description
In v1.1 i continued the training up to: 10145 steps and 34 epochs.
FAQ
Comments (26)
Well done. Does anyone know the prompt or technique to get a zoomed out view that shows the full body and not just the upper? I tried "Zoomed out" etc. and that doesn't work. Thanks!
i did that yesterday, just describe the butt, the legs and heels for example..
Hello, I have a question: how many repetitions did you use? and the batch at 1? or at 4?
Thanks for your feedback :) and nice work
Sampler DPMPP_2M works better than Euler
Great job!
Works well. If you want something specific, give v1.0 a try too, it seems more flexible which makes sense as v1.1 is the same but trained further. I suspect the training data included people with tattoos, getting quite a few tattooed people with v1.1. (After only limited testing, may need to play with strengths and prompts more)
Yeah, theres one girl that jave a big wings tatoo on her chest, oretry ugly if you ask me, but the blowjob she makes is epic, you might be right about the 1.0, atm im training again with the same videos, this time i split them all in 3sec duration clips so everything has the same length, FPS and resolution, lets see if It comes up better
Do you caption the tattoos? My understanding is that is important if you want to be able to prompt for no tattoos. (same with watermarks).
Seems heavy on tatoos and eye makeup, like a lot of loras here to be honest, not sure how that can be avoided. Too much face training!
That's because most of the trained on videos have similar looking women. The only way to really getting around it is training on hundreds of clips at lower lr, but it's too expensive.
@wiiz maybe is coz the captioning, i guess some words repeats in every caption file and that makes them to be mixed having the same characteristics?
@wiiz I will try and train a video lora but if it's anything like image based loras you caption out. Yes it may be harder to caption out faces but that's what style loras are so unlikely (i've just never done it before personally)
It 100% should not be hard to ensure tattoos are not in the train though, I mean you could just remove them in photoshop pretty easily if standards were higher here
I keep getting a white guy's arm with a very distinctive arm tattoo on it, so may be more than just captioning.
@KinkMaster can you point out that tatto on any of the posted videos? .. i didnt notice anything like that, i mean i didnt specifically caption for tatoos
@K3NK That's the whole point of captioning, you need to mention it so it DOESN'T train them.
Yeah I keep getting same faces and hoop earrings now even with character loras with 1.1
only three scenes has those earrings in the dataset videos, have you tried to prompt the earrings you want, or even "woman without earrings", but yeah, youre right, they are almost everywhere xD
you can just pre-generate a lower resolution version, than put it into V2V and increase the size, and lower the strenght of this lora while using the character lora you want, I'm pretty sure that will solve your problem
@liquidhead440 Nice to know, but that kinda takes forever.
@playnproto266 it really isn't , when you do video to video, you can just use really small size in the first step in text to video, like 336x208, and like 6 steps, my potato 4060Ti with 16Gb of ram does it I think around 40 seconds, the main point is to see somewhat what it would look like, because when you do step 2 which is vid to vid, it will look at it as noise, and nothing else, so you don't really need super good quality, just enough to kinda guess what you will get, I do this all the time, because I upload stuff to youtube, and it would take me ages to make a decent amount of videos without these steps LOL
@K3NK when you make a lora and do the captions always mention the model has earings etc that makes them optional to the model
@liquidhead440 Do you have a workflow you can share? I tried implementing something like this but the results were... not great. I'd like to see someone who knows what they're doing show us the way!