LoRA for two women performing POV oral sex on the viewer.
With one seed that gave me 2 penises, I put "multiple penises" in the negative prompt and it fixed it. I'm not sure if I just got lucky or if this is a general fix. It seems to happen a little less often than with Hunyuan, anyway.
The "Training Images" download contains a sample workflow.
Some phrases you can try, though it won't always listen perfectly:
Two women are {kneeling|leaning over} and positioned on either side of a man's erect penis.The woman on the left ...
The woman on the right ...She is moving her head up and down as she takes the penis into her mouthShe has her tongue out and is licking the shaft of the penisShe is licking the man's testiclesBoth women are sticking out their tongues and licking the shaft of the penisThe man's lower body is visible, showing his nude stomach and legs{The view is POV|The view is from a low POV angle|The view is POV overhead from above}Description
FAQ
Comments (10)
The model works very nicely in T2V, but in I2V it tends to modify faces.
Is it possible to get an I2V model trained with blurred faces, or something like that?
You can't really mask or blur faces with this because faces are so critical to the movement itself. Have you tried lowering the lora strength? I'm not sure how low you can go, but try to lower it as much as possible without ruining the motion.
@dtwr434 yes I've tried, but I can't find a sweet spot somehow. 0.8-0.9 seems better, but still it is quite unreliable. I mean, I am very happy with it, but it's like one small step from perfection :)
@dixawo7135618 Hmm, yeah I'm not really sure what to do about it. Any sort of tool to blur or mask faces would include the mouth area, and that would ruin the entire lora. The only real way to do this would be to partially mask the faces by hand for each frame of video, and I'm not going to do that as it would be tedious and take forever. Maybe eventually someone will figure out a better strategy.
One thought that someone presented to me recently for cases where you can't blur is to try to use rare tokens to name the people in the videos. So like "A video of Ovzh1 and UytI4 sucking a large erect penis..." or perhaps "A video of two women sucking a large erect penis, Ovzh1 is on the left and has..., Uytl4 is on the right and has..." etc. with the hopes that the likeness gets attached to the rare token and not the action. I haven't gotten around to trying it but it is a good idea in principal that might could be helpful. Other than that, just use as many varied faces as possible and don't repeat faces in the dataset is about the best you can do when you can't blur or mask.
Hmm, that is an interesting idea.
Alright, I need an idea to stop with the double penises. Anyone?
Have the same issue. You found a solution?
lost_moon Nope. Seed, I guess - or try 'the other sucks the same penis'
the more the better
Details
Files
wan_double_blowjob.safetensors
Mirrors
wan_double_blowjob.safetensors
wan_double_blowjob.safetensors
wan_double_blowjob.safetensors
wan_double_blowjob.safetensors
136_wan_double_blowjob.safetensors
wan_double_blowjob.safetensors
wan_double_blowjob.safetensors
wan_double_blowjob.safetensors
wan_double_blowjob.safetensors
wan_double_blowjob.safetensors
wan_double_blowjob.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.