This Lora is trained on the WAN 2.2 I2V 14B low and high noise models and can be used to generate facial cumshots. It works with or without a man present in the initial image.
Update 07/01 - Alt V2 out
Happy new year everyone! Hope you are all having a nice start of the year!
After some more testing, I decided to also put here a High/Low alt version of the V2, which seems to give better results in some situations. I think it's mostly in more zoomed in shots but could help in other cases as well. If there's something you are trying with the normal V2 and you think it could work a bit better, try out the alt version then.
Update 01/12 - New version out
Not everything I was going for worked out with this version, I need to tweak the dataset a bit a train again but I was anyway happy with the results and decided to share it.
Some concepts are not quite there and there's quite often alot of cum but... Enjoy!
Please check the Training Data zip attached to the High version for all the captions used in the training. Videos have the workflow attached.
Update 02/10 - Attached workflow file (to the high version)
Trigger Phrases (some work better than others, the least tested ones are highlighted below):
a woman is kneeling in front of a man
a woman is lying down on her back and the man is on top of her
a woman is kneeling in front of two men
cum shoots out of their penises and lands on her face and eyes
(cum shoots out of the penis and lands on her tongue)
cum shoots out of the penis and lands on her face and her eyes and her mouth
(cum shoots out of the penis and lands on her face and her mouth)
cum shoots out of the penis and into her mouth and tongue
big cum load
(high speed cumshot)
(thick cum load)
(cum strings)
she has her mouth open
she has her mouth open and her tongue out
she has her mouth and eyes closed
(she nervously smiles)
(she smiles)
(she flinches)
she frowns, she hates cum
she hates cum
the man strokes his penis back and forth with his hand
the men are stroking their penises back and forth with their hands
the woman strokes his penis back and forth with her hand
the view is from above
the view is from the side
big close-up
close-up
Works pretty ok with 6 steps and Wan21_I2V_14B_lightx2v_cfg_step_distill_lora_rank64.safetensors at strength 3 (high) and 1.5 (low).
CFG 1 - Euler Ancestral
Description
FAQ
Comments (63)
recommendation: do not use "shoot" or "pulse". use "ejaculates" because that's what it is and it won't cross-infer something else.
i can certainly try it out for next time, thanks
Is this a T2V lora? I'm using it on I2V, but I often get strange results, like a man's face with a penis in his mouth or ejaculating through his fingers.
This is I2V but yes, unfortunately that also happens :(
It's a little stubborn sometimes, but depending on the overall composition, something along the lines of "you can only see the man's lower body and his erect penis poking out of his jeans" seems fairly effective.
After extensive testing, I have concluded that this lora does not work well with some images. For some reason, it is sometimes unable to render the hand that is holding the penis what causes the penis to be deformed. With other images and the same prompt, it works great. It's a shame it's not always like that.
So far I really like the results. My GPU is running non-stop since this morning churning out cumshot after cumshot. While some are absolutely great, others have the usual quirks like disappearing cum or a penis where the belly-button is etc. But the LoRA works fantastic when a penis is already in frame and the fact that the LoRA almost always finds a way to let a guy walk in an blast his load is just 👍🏻.
Speaking of walking in: were there people randomly walking through the frame in the training data? I swear, as soon as a door or some kind of realistic background is in the image, a random guy shows up and walks through the video.
I’ve posted some examples. It’s hilarious.
ahahah :D actually no, none of the training material has a guy walking in, that's all Wan I guess :)
It's actually one of the improvement points for the next version.
Regarding disappearing cum, I mainly notice it in videos longer than 5 seconds, not sure if you have the same experience.
You can actually chain two videos together and in the second one if you just say the she has cum on her face ( and its flowing down her face or something ), it will still keep animating the cum.
This one is really impressive, oh the messes I’ve made with this.
high praise, indeed.
Thank you :)
Definitely the best looking of the "cum model".
There does however seem to be a thing where cum appears at odd times, like if doing a facial sometimes the object has cum coming out of the mouth before anything cum is shot. Maybe it can be fixed in prompting, not exactly sure how though.
Sometimes it can happen unfortunately. Rerunning the generation again can quite often give a result where this doesn't happen
Best Wan 2.2 Facial Lora out right now. Great work!
This is good, but is there a way to stop the person from opening their mouth? Even when I write prompts for mouths and expressions, they usually open their mouth.
you can try: she has her mouth and eyes closed .
@Civitaiwolfxx Thanks, I'll give it a try!
How do you use the multi prompt layout and combine video options? I've been trying to make a complete 24 second video using all four prompts but I haven't been able to get it to work.
You have to enable the prompt group and the subgraph next to each prompt group. then enable the video combine node corresponding to how many prompts you are using. You might need to check inside the subgraphs though, I'm not sure they are configured for this specific lora.
@Civitaiwolfxx thank you so much! that was it! i needed to check in the subgroups to change my model. i have it saved under a different directory name.
This is pretty awesome lora. Thank you for it!
I'm pretty new to comfyui, but on that included workflow, I was able to generate some nice results, but I can't seem to get the upscaler enabled, is that broken on the workflow or I am just missing something? I didn't change any settings except the loaded image and path to the various files that I didn't have a wan subfolder for. On the Load Upscaler it can see them.
I haven't used the upscaler there in a while but should working if you have upscaler models downloaded. Did you enable the Load Upscaler model and the Upscale Image (using model) nodes ?
@Civitaiwolfxx I must not be seeing the option to enable them, both the Load Upscaler and Upscale Image nodes show disabled, I can manually force them on from right click properties, but that just seems to save a non scaled file. I'll fiddle and see if I can't get it working.
I guess when they are disabled like that, the option to enable from right click is called Bypass. Like i said, I haven't tried it in a while so I'll try it in a bit and let you know.
I attached a new generation just now that was done with upscaling and it has the workflow in the video. you can try that one
@Civitaiwolfxx Oh, I am an idiot. I was using the workflow on the v1 download. Testing the one from your recent upload. thank you so much!
@Civitaiwolfxx I finally got it working with the updated workflow. Had tons of issues getting the manager to install all the dependencies, but it is working! Thank you for taking the time to reply! ❤️
@NeuroSentari Glad to hear you got it running
folloed all the instuctions but the lora doesnt seem to want to work, always get a weird half a chicken foot looking penis and no cum, anyone got any idea what I'm fucking up
do you have the Light2x lora as well ?
I am getting sausages (literally) instead of penises and milk (literally white liquid) while using this lora. Tried the trigger words but no help. I won't be using it again any time soon, seems f4c3spl4sh is way more consistent. This is not image generation, this takes time, even more on an 3080; I can't keep wasting time and electricity bills to get failed clips.
sorry to hear that. I can only say that those sort of results I usually saw when I was using the High model in the Low pass.
@Civitaiwolfxx it can also happen due to speed up loras and / or the combination of them.
@Viniblo lowering the strength can help
@Lora_Addict Sure but if he's using my workflow (no clue) then it would more likely be what I mentioned. but thanks for jumping in and supporting :)
@Civitaiwolfxx you are absolutely right, i always use my own WF so i was projecting onto others, my bad :D
Can't load the workflow from the posted videos, are they supposed to be .mp4? is there something I'm missing?
They are supposed to be mp4 indeed i guess. Just drag and drop it to Comfyui. either from the first post or from my latest one. Some of the in between posts didn't have workflow
Sorry for going off topic, but are there any Loras for Wan 2.2 that don't cum up on the face or body without stretching the penis and other deformities?
Not sure, at least I didn't have any body cumshot samples in the training data. Maybe someone else can answer that
Probably not, Wan dicks are weird. I've made a lot of Wan loras and a lot of the time the penis ends up looking like a smooth meat tube.
@TwoMoreLurker true enough
I have a problem: the girl always puts the penis in her mouth. How can I make it so it's just ejaculation without the penis in her mouth?
can't say i experienced that. does it happen with the prompts in the examples for instance ?
First time doing this. Am I supposed to run both the high and low V2.0 versions of this at the same time? Or only one? If both what strength for each one? If just one same question.
Hey there! Yes, you need to run both the high and low V2.0 versions at the same time. Load them independently, with a KSampler for each one (that’s how WAN 2.2 works). I’m a beginner, but this worked for me. :)
sorry, was away for xmas. yes, both versions as @brokenmalas mentioned (thanks @brokenmalas for stepping in :) ). Strength is 1 for both. Check my workflow for examples
Any tips for keeping the original face? I would like to use this on older women but each of these facial loras removes wrinkles or unique facial traits and makes the woman's face smooth.
I will need to experiment that scenario before commenting
there is a lora that helps with older women. It can help with the skin texture, but if you are trying to keep a specific face it won't help with that. It's T2V but works well with I2v. Also turn the strength down of this cum lora https://civitai.com/models/2266068/mature-body-realism-lora-female
@Kiefstorm Thanks. That actually helps not smooth out whatever skin is still visible.
Says its missing: wan22_realcum_high_noise_v1 and low noise cant find them anywhere.
added them manually instead, but my issue is, running 10 seconds, it just runs the first 5 seconds twice to make it 10 seconds, any ideas?
i run 10s, 720p, single image to video, using 2x upscaling, default 16 fps interpolated to 32, its basically just giving me two 5 second clips in one.
so as an alternative i tried using the additional prompt entries, but they give me this error: FileNotFoundError: Could not find module 'nvrtc64_120_0.dll' (or one of its dependencies). Try using the full path with constructor syntax.
Isnt that version super old? Im on 13.1 so that file doesnt exist on my PC.
I've never tried 10 seconds, not sure the workflow would work for that but I mean, this is trained on up to 5 second clips, it would not really work well for generations much longer than that. split it into 2 second generations
@Civitaiwolfxx well my current issue is I cant use the extra prompts without getting that issue about some nvidia dll, seems related to interpolation? based off google searching, but the first step is fine all the way through the end, and second and third and fourth are just the exact same versions as the main workflow no? So I dont get what the issue is. Also that dll seems really old. (Im using your workflows attached to the mp4s)
If its related to interpolation, maybe you can try disabling the RIFE node
@Civitaiwolfxx But the thing is that the RIFE works just fine if I just run a single prompt run. Its when I enable the second one for another 5 second clip, that it shits out that error. Is there some oddities inside the hidden prompt thats not the same as the regular workflow?
@hamzta yeah i use wan 2.2 wan2gp via ponokio and i have see same phenomenon happen there if i want to continue a video. I have to turn rife interpolation off so both clips run at 16fps or the video gets wonky for the second video. so i just do everything at 16fps then use topaz labs for upscaling and framerate interpolation. also topaz ai's framerate conversion doesn't have that weird artifacting like rife does. I only use rife if I know I'm only doing a 6 second shot and moving on to the next. good luck and Cheers!
can anyone help plz. i'm new to all this. got comfyui installed, lora and diffusion models, when i import the workflow is get errors:
"workflow has missing nodes"
JWIntegerDiv
FinalFrameSelector
PainterI2V
JWFloatToInteger
JWImageResizeByLongerSide
Int
custom nodes missing. and somehow does not install in the manager? anyone?
the painteri2v you don't really need. the others i think are from : comfyui-various suite
"comfyui-various suite" is not listed in the manager.
https://github.com/jamesWalker55/comfyui-various or just replace them for something else. the resize one just replace with a standard resize node. the other ones are just doing math so you can also replace it with something else or just add the result of the math there by hand. I guess they are dividing the steps in half
got the node issue fixed by installing comfyui and manager manually with this guide:
https://www.stablediffusiontutorials.com/2025/08/wan-2.2-video-generation.html
Details
Files
Wan22_CumV2_High.safetensors
Mirrors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
high.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Facial_HIGH.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
103_Wan22_Cum_High_dec06.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Cum_Facial_v2__i2v_HIGH_.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
fp-high.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CivitaiWolf_Facial_v2_HIGH.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
facial_wan_high.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
wan2.2-i2v-high-cumhelperV2.safetensors
Wan22_CumV2_High.safetensors
Wan22_BJ_CmV2_High.safetensors
C_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
test-52-high.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
8jCrUYsG.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
high_Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Wan22_CumV2_High.safetensors
Facial_HIGH.safetensors
Wan22_CumV2_High.safetensors
Wan22_CV2_High.safetensors
Cum Facial v2.0 I2V HIGH - Wan2.2 - she closed her mouth and looks up, cum shoots out of the penis in twin jets and lands on her face and clothes and into her mouth, she reacts to the cum by closing her eyes and opening h.safetensors
Cum Facial v2.0 I2V HIGH - Wan2.2 - she closed her mouth and looks up, cum shoots out of the penis in twin jets and lands on her face and clothes and into her mouth, she reacts to the cum by closing her eyes and opening h.safetensors