For Wan2.2:
I know, I know, there are other models similar to this, why bother? Well, I like to have my own models for the basic porn concepts, thought of sharing with the community.
Trained on 22 clips of 3 seconds, 16fps, 512px*512px.
Use this
A woman is receiving a load of cum on her face, a man holding his penis in his hand and stroking it ejaculates cum or white semen into the woman's face.Then customize to your liking.
Workflow is included in the samples.
If you enjoy these models and find them useful, please leave a thumbs up. It really motivates me to keep creating more! Feel free to also share some of the work you generated with them in this page; I’d love to see it.
Enjoy!
For Wan2.1:
This LoRA spits cum on faces.
Works great with character LoRAs.
Some of the clip samples are interpolated from 16 to 24 frames, thus the slower motion.
Start with the base
wmgtnfcl, A woman is getting cum on her face from a man's penis. Then customize to your liking.
Trained on 22 videos 256*256 between 2 to 3 seconds.
Enjoy!
Description
FAQ
Comments (48)
how is it's performance on I2v, and does it effect face consistency?
Nice work, but do you know why mine oversaturates every output? its like the color is turned it too high...
What number of steps and CFG are you using? Follow the matadata found in one of my videos and let me know.
@jm112368767 have you figured it out? mine is too saturated also
Could be conflicting with other loras, could also just be "too strong" try lowering the lora strength, also like f1985 said if your cfg is not optimal your colors can get weird, also if not enough steps it can bone you too. Video generators work in much the same way as image generators, in as much as how they respond to stuff like lora strength, cfg and steps. Also, I haven't tried it yet but I'm going to be experimenting with WAN using some fp16 text encoders and see if I can get something to work without having to re-code something. I've been looking and so far, I can't find anybody who's using fp16 clips/encoders, etc. my whole image pipe is fp16 and there is definitely a noticeable difference in quality and prompt adherence. So that's my quest these days, if anybody knows of a resource out already where you can do a full fp16 wan2.1 I2V pipeline that would save me TONS of work. lol
Does the image need to have the presence of a penis in advance?
I didn't train this for I2V, let me know if that would work with you or not :D
@f1985 it does actually! you just need a "penis" wan for lora.. Im testing right right!
I have tried using pictures with a penis, but I have not tried using pictures without a penis. It takes a long time for my machine to run once. If anyone has tried, please leave a message
@ggw19880307 I just posted a new video which I did
@dxjaymz 哈哈哈，阴茎已经离开身体了
I hope you make an i2v version
Seems like it does work with i2v after all. some of the guys have already posted to the gallery :D
If a lora is trained on T2V it is by nature good to go for image to video I've heard! So good job op! Might wanna try some videos from spermmania maybe? They got the perfect setting for it :D
yeah, no need to retrain this. works like a charm with i2v
This is one of the best I've seen. Do you mind posting more about your training setup? Like LR, repeats, epochs and all that?
Kept everything at default except the repeats, I crank that up to 10 when i have small dataset, and I save every 5 to 10 epoches during trainings and test each file on the spot, see if the lora can do what is expected to do first, then if it can deliver when combined with a character lora (different character LoRAs to count in for the difference in their training settings).
I am newbie at this, so I apologize if I am not explaining things well.
@f1985 Newbie? You're doing a great job. Are you using diffusion-pipe for training?
Also, are you training your own character loras? The ones on Civit for WAN are not that good, as compared to Hunyuan IMO.
@rollpollution193 thank you, yes i am using diffusion-pipe. i trained my own character LoRAs as well using diffusion-pipe, until now I've noticed that a smaller dataset, even with high repeats, isn't as good as when using a large dataset.
@f1985 Interesting. What is the size of the dataset, and repeats you're using? Also are all the other settings the defaults like you said?
@rollpollution193 Trained on 22 videos 256*256 between 2 to 3 seconds. Repeats at 7. Everything else is default.
@f1985 how many frames buckets did you use for each clip? I am trying to train a motion lora myself but I'm kind of stumped on that part. Some say 6 frame buckets is enough, but I'm unsure if enough of the motions would be captured from just 6 frame buckets, for example [1, 15, 29, 44, 58, 72]. Did you use more or just the default which I think is 3 frame buckets?
@SLAPY7 Default.
contrary to what other comments are saying, i dont think this works for i2v. no generations at all ever have any shots. would be great if it did work or if there are any pointers
Try using longer prompts, get an image of a cumshot and get its prompt, and use it along with the trigger words.
Same here. I even used images with a huge cumshot already there: it disappears. And then nothing more. Gotta couple that with other LoRAs to make it "work". Aka at the end I managed to get results, but it's probably more thanks to other cumshots LoRA than this one. Trained on t2v usually LoRAs work fine on i2V, but cum is a quite complex concept to generate. It might need a separate i2v training.
its working very good for me.. I'll post some vids soon after I render in higher resolution (you need a "penis" LORA) which I got from Wan 2.1 Penis. Edit: I think the post processing of the videos are removving the cum (watching the creation now I see cum on the first pass only)
@dxjaymz @SD_AI_2025 I've had moderate luck when doing a 1 frame with t2v first an, then i2v using that frame once I get a frame I like. But it's still not gushing out like in the samples, just a faint shot for the first few frames. T2v works a bit better. Again I've yet to see like the samples,I'm not doubting it I'm just having a harder time I guess.
I ran this by itself on I2v and also alongside other loras and it does work well. SOMETIMES you have to prompt that the camera stays still because sometimes it will spit out a vid with CRAZY camera zooms or flyovers, but even with that said, easy to prompt-out. It also works with the other jizzlobbing Wan lora quite well but it does take some dialing in... its REALLY easy to overdo it on both, then over-compensate and dial them too far out. But if you can hit that sweet spot the final results are pretty good. I'm going to make a lora for this type of thing as well the next time I sit on an h100 for a few days.
Thank you. Excited to see the results.
Just to get it right, for slow motion the 24 fps but for "true" speed, I should convert my training dataset videos to 16 fps because that's what Wan 2.1 use?
In general what would you recommend for the dataset videos FPS for training LoRA for Wan 2.1 i2v ?
I'm having trouble getting this LoRA to work correctly. Most of the time I end up with a mess (no pun intended) - or just way too much cum. Does anyone have any suggestions as to the best settings to get good results (shift, cfg, model strength?) For that matter, shouldn't this work with Skyreels (since it's based on Wan) as well? At the moment I've mostly tried using it with the diffusion forcing version of Skyreels 540p.
Hey everyone.
You can refer to my work, but most of it is animation, so it can have a dense or rich feel. I think prompt words are very important.
Here are the three LORAs I frequently use, along with their strengths.
These three are always used together.
strength:
Wan Cumshot : 1
Facials - Wan2.1 T2V 14B : 0.7
Wan 2.1 Facials - Cum on face, Bukkake, Lora, T2V, I2V : 0.5
----------------------------------
Regarding "wan cumshot": From my perspective, this term provides the trajectory of the shot. However, the downside is it can sometimes result in a missed shot, unless you start very close to the face. Personally, I prefer a shot from a bit further away.
For "facials": This clearly specifies the target location: the face.
As for "cum on face": My personal experience with this term is that it conveys a sense of large quantity and thickness.
I rarely use real people in my creations, though.
---------------------------------------
For real people, I'd suggest sticking to these two for the best results. You can also add Detailz-Wan: Detail Enhancer for WAN Videos.
Here are the settings I'd recommend trying:
Wan Cumshot: 1
Facials - Wan2.1 T2V 14B: 0.7
Detailz-Wan: Detail Enhancer for WAN Videos: 1
maybe add penis lora , you can choice
Thanks for sharing the tips for the 2.1 lora. Any tips for the 2.2 version you don't mind sharing? Thanks in advance.
Hope you can update this fantastic LoRA to use 'Wan Video 14B i2v 720p' base model since the latest generator changes.
Please update.
UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE UPDATE
why can't i select it?
This is the only good facial/bukkake lora out there and it's unusable. fuck
"Create" does not work. The resource does not show up when trying to collect the various LoRas for my video lcip.
I will literally hire a private detective just to track this dude down so he can update this LoRa.
Updated ;)
@f1985 so, did that detective leave you alone yet? 😂
why mouth / togue are not touched by cum?
Because nearly no model does that yet. Tongue changes shape constantly, add to that the camera angle and you've got basically an infinity of "tongue" shapes. Hard concept for an AI to fully understand yet. Add to that a fluid that sticks to the tongue and it complexifies even more the concept. Models still struggle with that, understandable. What we get now is immensely better than what we had a year ago.
Just gotta be patient.
very good
Details
Files
facials60.safetensors
Mirrors
facials60.safetensors
facials60.safetensors
facials60.safetensors
facials60.safetensors
wan_facials60.safetensors
facials60.safetensors
facials60.safetensors
facials60.safetensors
facials60.safetensors
facials60.safetensors
facials60.safetensors
facials60.safetensors
facials60.safetensors
facials60.safetensors
facials60.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.