Could you share any examples, so we can see the problem? The examples here don't seem to do what you say, so I want to see how it happens with normal use.

@Jellai For example, if I apply Lora to a photo of an Asian, it will turn into a Western face.

@Bangdes Oh yeah, I can definitely see that happening based on the results I see here. Okay. Thanks for clarifying.

@K3NK Thx very much. I'll try it.

@K3NK will this work on 2.1?

10,000 it's expensive for both models, so I'm going to wait

The faces get smoothed out a lot. For example wrinkles or skin marks vanish.

@Bangdes I think that's basically all of his lora from my test and you need to tweak it a lot and 1 out of 10 tries you get a somewhat similar face.

Facial deformation can be eliminated using https://github.com/shootthesound/comfyUI-Realtime-Lora
Disabling specific lora layers removes the issue completely

@_qwerty_ Which specific lora layer removal worked for you?

@Jellai for HN, late only except "other weights" and for LN, disabled first 2 and the last 3. But I'm still experinenting

Creators deserve credit.
But 10K for a lora? Come on...

@vbnfghn - you don't HAVE to pay for it. You could just wait 14 days and get it for free.

looks a little weak, you can get better with high noise cumshot aesthetics and another cum lora together.

@oldthrashbar - I haven't had a chance to try it yet, but I hope it's not weak (or v2 comes out quickly). This is a LoRA I have been looking for because I was unable to get the desired affect. There are many cumshot LoRAs out there, and I feel like I've tried them all, and I have never been able to get more than a regular facial. If you have examples of better results with other LoRAs, I would love to see them, if you can share?

@HibsMax I will

yeah a lot of things about this site are completely fucked.

But fellow gooners remember: Gauge the probability a lora is being deleted before the early access period runs out and decide accordingly.

Low Risk Lora:

Buy it now using buzz (and support the creator), or just wait a couple of weeks

High risk / dangerous Lora likely to be deleted:

Better shell out some buzz or you might never see the lora again

Nudify / Nude scanner / Titty Flash / undress loras , and also deepfake utilities get deleted quite regularly and almost every time.

You decide but don't work yourself up.

someone message me when its free

@oldthrashbar So the high noise is the one to increase to get a better outcome from this LoRA? Not both? There are a few spectacular uploads in the gallery that I'd love to replicate.

@SmutSurfer1960 far as i understand it, high noise dictates general theme and motion, while low noise fills in the details. I think of high noise as the well meaning but unskilled apprentice and low noise as the master that cleans up after him and makes sure the job is done properly. lets take a BJ lora for example. using only high noise you will get a BJ scene, but the penis and details just in general will look off. slap on the low noise and suddenly its correctly textured and divided into glans and shaft and balls and not the monstrosity from before.

I'm not generating videos, yet. Although, I noticed that whenever what I prompt doesn't work, it's because the lora is strong enough to restrain my prompting. Have you tried lowering the lora's weight gradually while adding you prompt? I'd start there if I were you. Feels bad OP didn't even answer you 🙄. Keep up!

@GlowingGuardianGirl Thanks but yes I've tried different strengths and so far it seems pretty random. Sometimes the output, of cum, is decent and other times it's a dribble.

@SmutSurfer1960 I read on different comments that prompts like sticky, thick, accumulation, filling etc can help getting better cum. I'll link if I ever find where I read it again. Hope this helps

he is a man of focus and sheer fucking will

as an Ancient Wakandan proverb says "Everything cums to him who waits..."

use this low lora at 0.5 then use the cum v2 at str 1 on high and low. worked for me :)

Cum/Facial - Wan2.2 - v2.0 - High | Wan Video LoRA | Civitai

@sirbazzrick appreciate it will give it a go, the mouthful lora is great just lacking the cum for whatever reason lmao!

@K3NK do you mean the lightning loras? bit confused as to what you mean by speed loras mate, loras got so much potential for what I want to do with it just can't seem to yield any more cum that like 2 drops really frustrating rendered like 50 iterations of strengths and samplers to no avail so far, even the cumfacial lora hasn't helped feels like something is "blocking" the desired out come, the experiment continues.... ( I forgot to mention I'm trying to do 12v 14b) if that matters

this lora for sure as already recommended keep at str 0.5 both high and low, u will get movement and there will be no face change, it does plastic cosmetic surgery to your character at STR 1, it's just to funny to see transition and change to skin, lips and nose remodeling at str1 :D but some folks might not mind that change and u can use the cum v2, or us can go with F4C3SPLASH at str 1 or WAN 2.2 Anime Cumshot Aesthetics as well (this one works very well at filling mouth with cum and cum stays on tongue or mouth)...

@K3NK i know you can make things better and better. your loras are phenomenal.

high on 0.5.- 0.6 does not change faces, low can remain on 1.0 , just use workflow with WAN MOE K sampler, lightning 4 step Loras with high set to 0.5 and low to 1.0, go with 8 steps with such setup and bump cfg in WAN MOE K sampler to 3.5 for both high and low. for testing use use 576x768 res for quicker render, u will get results on par with what u see in the sample videos...I for sure did...with high lighting lora on 1.95 and 8 steps and cfg 1 on both high and low u can get similar results as in samples but cum will be less thick and more fluid actually kind of more realistic. 1.95 on high can break motion in some loras but with this one just works ok.

@K3NK it's always a guessing game with this shit... One thing changes, everything changes.

haven't had one face change. stays almost 100% the same. Used a very simple workflow and lightning lora's

If you are using an I2V workflow you need to use an I2V model. Try the Dasiwa or Smooth models.

My example with the milk bottle was a full T2V workflow and model.

I've used several models including smooth and I'm getting the same thing. Turns into a different person no matter what

these videos have the workflow and you open in ComfyUI https://civitai.com/posts/25725230

https://civitai.com/images/113321197

Usually if this is happening a lot it is because the model is being given WAY too much time as far as steps and WAY too much pressure (really cranked Lightning/speed LoRA) to make decisions. when you combine these things (too many steps, too much pressure) the model will go out of its way to find things to do to 'make the sampler stop yelling' even if it had already decided once and was ready to hand off to low noise. When this happens, decisions that were already made about identity, pose, etc end up getting reopened and reevaluated and this is how you get changing identity, etc. The model during sampling doesn't decide once and then sit back and spend the rest of compute on making things look pretty no matter your sampling schedule, if that was the case you could just set as many steps as you were willing to wait for and get a better video. Please provide a workflow and models you have tried if you can. I am interested to see what seems to be causing this, as a lot of people comment in many LoRA about this. Are you using a workflow that uses the PainterI2V node by chance?

@whatthedeuce  yes i use i2v with i2v models, and thank you for your reply. i will Try the Dasiwa or Smooth models

@zodoor694 noted, thank you. are you using i2v like i am? 

@whatthedeuce perfect thank you, i will try this 

@clzpetn804 this is the workflow im using,  https://civitai.com/models/2050408/fatbergslim-image-2-video-workflows?modelVersionId=2430844, and thank you for your detail explanation.

@whatthedeuce YES!!!, works perfectly, thank you so much!!

not here. works 100%

what model and workflow are you using for I2V? Are you using the PainterI2V node by chance?

In my case, no change. It slightly deforms the shape, because of a extreme close up, since it simulates the camera angular lens. But the face stayed the same.

For me it wasn't changing the face. then i changed something and it now does change the face. Im confused cuz i changed it back and now it still changes the face.

Face does change for me, subtle changes, such as eyes and eyelashes etc... weird.

@treeofshabadoo636 @Erosire something I should be asking and we should clarify: are we talking about changing the face over multiple videos or in the same video? this is a really key distinction

@clzpetn804 same video. maybe 2-3 secs in it changes to a featureset averaged out between maybe five girls, likely from the training data