This is my newest and most revolutionary version of this lora. I realized most of these lora's focused on one particular angle like pov, from the side, from the front and thought why can't one lora do them all? So here you go, version 3 of my lora.
The relevant keywords are "under view", "side view", "frontal view", and "pov" and of course doggystyle. They each control the preferred angle of the output.
"Side view" is the traditional view from the past versions.
"frontal view" is from a forward angle. Not directly head on, at a slight angle but with the woman facing almost directly towards you.
"pov" is exactly what it says. A direct pov shot.
"under view" shows you from behind and under so you get full penetration view with full dick and balls in view. The only problem is that because Wan struggles with genitals you sometimes get weird genitals but they can be improved by adding a second lora that focuses on genitals.
Example prompts are:
an under view of a woman with red hair being fucked doggystyle as she faces towards the left. Her ass is in the air with her vagina visible as a man with a large black penis begins thrusting, his penis visible as it slides in and out of her. She is bent over a desk and they are in an office setting.
A side view of a woman with pink hair and pale skin being fucked doggystyle as she faces towards the right. She is positioned with her face on the ground and her ass in the air with her hand on her ass as a large white man is positioned behind her thrusting into her with his large penis visible as it slides into her. They are on a beach and she is face-down in the sand.
A frontal view of an african woman with green hair and dark skin being fucked doggystyle as she faces towards the right. She is laying flat on her stomach as a large asian man is thrusting against her, his large penis visible as it slides in and out of her. They are on a public sidewalk in a busy city with onlookers watching.
A frontal view of an Asian woman with black hair being fucked doggystyle as she faces towards the left. She is on her knees, with both her arms held behind her with her chest pushed out so her breasts jiggle and sway beneath her, Behind her an asian man is holding her arms as he thrusts against her. They are on the moon with the earth visible in the background.
A pov of a woman with brown hair and white skin being fucked doggystyle. There is a red strap going across her back, and her ass is aimed towards the viewer. Her anus and vagina are visible, and a large white penis is sliding in and out of her vagina as she pushes her body back and forth against it.
You can prompt facing left or right to pick which way the woman is facing as well. I trained it on realistic videos, but you can use i2v to use it for anime or whatever else.
Description
This is my newest and most revolutionary version of this lora. I realized most of these lora's focused on one particular angle like pov, from the side, from the front and thought why can't one lora do them all? So here you go, version 3 of my lora.
The relevant keywords are "under view", "side view", "frontal view", and "pov" and of course doggystyle. They each control the preferred angle of the output.
"Side view" is the traditional view from the past versions.
"frontal view" is from a forward angle. Not directly head on, at a slight angle but with the woman facing almost directly towards you.
"pov" is exactly what it says. A direct pov shot.
"under view" shows you from behind and under so you get full penetration view with full dick and balls in view. The only problem is that because Wan struggles with genitals you sometimes get weird genitals but they can be improved by adding a second lora that focuses on genitals.
Example prompts are:
an under view of a woman with red hair being fucked doggystyle as she faces towards the left. Her ass is in the air with her vagina visible as a man with a large black penis begins thrusting, his penis visible as it slides in and out of her. She is bent over a desk and they are in an office setting.
A side view of a woman with pink hair and pale skin being fucked doggystyle as she faces towards the right. She is positioned with her face on the ground and her ass in the air with her hand on her ass as a large white man is positioned behind her thrusting into her with his large penis visible as it slides into her. They are on a beach and she is face-down in the sand.
A frontal view of an african woman with green hair and dark skin being fucked doggystyle as she faces towards the right. She is laying flat on her stomach as a large asian man is thrusting against her, his large penis visible as it slides in and out of her. They are on a public sidewalk in a busy city with onlookers watching.
A frontal view of an Asian woman with black hair being fucked doggystyle as she faces towards the left. She is on her knees, with both her arms held behind her with her chest pushed out so her breasts jiggle and sway beneath her, Behind her an asian man is holding her arms as he thrusts against her. They are on the moon with the earth visible in the background.
A pov of a woman with brown hair and white skin being fucked doggystyle. There is a red strap going across her back, and her ass is aimed towards the viewer. Her anus and vagina are visible, and a large white penis is sliding in and out of her vagina as she pushes her body back and forth against it.
You can prompt facing left or right to pick which way the woman is facing as well. I trained it on realistic videos, but you can use i2v to use it for anime or whatever else.
FAQ
Comments (53)
V3 is really nice =) having all those poses and angles is one Lora is pretty cool!
GREAT WORK!!! 👏👏👏
I just tried to combine it with my doggy hairpulling lora https://civitai.com/models/1414532/wan-21-hairpulling-doggystyle . If you put my lora at low strength it works together pretty well =)
Yeah mine has a couple videos trained into it with hair pulling, so stacked would work pretty great I'd think.
This is so effing good. Thank you very much.
I'm shameless, but could you please consider making a mating press?
Have you tried just i2v with an image from that angle and just a missionary lora at low weight? Anything to help it with the thrusting movement
@asdrabael I tried using a missionary LoRA, but the results weren’t very good. The missionary pose tends to move sideways rather than up and down. The POV version didn’t show much improvement either.
Bim shala bim: Mating Press (Wan) - v1.0 (T2V) | Wan Video LoRA | Civitai
@chesthole Yo, I'm already using it lol, but thanks!
I'm a newb. Loving this but I'm trying to get the woman in the videos to stop opening her mouth. I've tried in several prompts but I wondered if clip/model strengths would help. The motions added are great. Thx!
Yeah, not having them moving their mouth isn't really possible. It's an issue with lots of lora, and in my training material the women were frequently moaning or saying stuff during the sex so they move their mouths a lot.
If I had changed up my captioning a little to differentiate between open mouth and closed mouth it might be possible but unfortunately I didn't.
Prompt wise how and where do we suggest clean bounce physics for breasts? Also can you post workflow? I am having a hard time finding a good one for the doggystyle loras. Thanks excellent work
These were done in the basic kijais t2v workflow with the lora node attached. Just put "breasts sway and jiggle" in the prompt to make them bounce around.
You can download the videos and they're also workflow embedded.
New best doggystyle. Amazing job.
I love your lora. The only issue I have is that it causes desaturation when I use it in i2v. As in, the colors turn desaturated. The generated videos are also prone to slowmotion, what framerate did your training videos use? Higher than 16fps?
Yea, the videos ranged from 24fps to a handful that got up to 60fps. I noticed it too after it was done and figured it's not a big enough deal to redo over because of the time this particular lora took (several days of training). It's faster to just fix it with a frame interpolation workflow if the video is good otherwise. Next time I'll redo everythings fps as part of prepping the data.
Not sure about the desaturation and how to deal with that.
How do i do this with just a picture
@asdrabael Ah, so that's why. Framerate needs to be standardized across the training otherwise one will get uncontrollable slow motion or fast motion. The frame interpolation workflow can solve it only sometimes. Let me give you an example. Let's say you give the i2v a picture of a girl squatting and you say "she slowly stands up, with difficulty,". If the video ends up getting produced in slow motion, in 5 seconds of generation, she still wouldn't have stood up halfway, let alone fully. You can't interpolate frames that don't exist. You'd then have to try extending that video with another run, starting from where the last stopped, but then you'd lose the context and be prone to getting inconsistencies.
Now, clearly, this lora isn't meant to be used for girls standing up, that was just an example. A more accurate problem for this lora would be that when slow motion triggers it is difficult to get a full in and out to the base thrust. Which becomes even harder if the rod in question is long.
Anyway, it's good that we know the problem now so when you do further lora training it can be avoided. 👍
As for desaturation, it is hard for me to say without seeing the dataset and captioning. But if I were to guess, your dataset probably mainly consists of realistic footage with toned down colors and due to you not mentioning the relevant keywords in the captioning, the lora associated not only the doggy position but the render style with it as well.
@Remiel
My dataset was all realistic footage, just standard porn colors except like 3 videos I got from a dataset given to me. I literally just went and ripped porn videos from websites and separated out clips of relevant sections and captioned them as realistic and focused more on directions and body angles than colors, lighting, etc.
@asdrabael Standard porn often has much duller colors than other types of videos, especially if it is amateur porn. You can try describing it as such in the captions or/and adding a couple of 3D or anime style videos to the training data to stabilize it and tell it not to associate dull colors with what you are trying to teach it.
How do i use on my own image. Cant seem to get a guy in there
Not sure what you're asking. Are you trying to make 2 guys with you in it or just trying to put yourself as the guy
@asdrabael i mean i use a picture of a woman. But how to make a guy appear and doggystyle her
@toiletsituation158531
Well I didn't train on any videos where the guy walks into frame but you could try an image of a woman bent over and using i2v caption something like "a side-view angle of a woman being fucked doggystyle after a man walks in from out of the frame and penetrates her from behind" or something similar.
Hey buddy I am new here, I know how to generate images with checkpoints and loras but dont know how to do those videos. Could you specify the steps and I will do it. Wan video and lora.. Also don't know which workflow to use... Thanks man
I use the WanVideo Wrapper custom node that is made by Kijai, and he has 2 workflows in his example folder, a t2v and an i2v. My example images were made with t2v but most of the videos people are posted are made with i2v. The lora is well, this lora the v3 model.
I've tried some T2V and they were all too complex and crashy, until i found this one named : Native_ComfyUIWan_Video_Text2Video-Upscaling_FrameInterpolation.json , along with diffusion model : onTHEFLYWanAIWan21VideoModel_kijaiWan21T2V14B.safetensors , clip model : onTHEFLYWanAIWan21VideoModel_kijaiUmt5XxlEncWan.safetensors and vae model : onTHEFLYWanAIWan21VideoModel_comfyorgWan21Vae.safetensors , all 3 available on the same Civitai onTHEFLY page (look for it). It allows you to add your personal character loras in addition to other loras. I trained my personal lora using Civitai's onsite trainer to about 2k steps and the result is jaw dropping.
Does anyone know why it is not usable on site?
Just wondering how to get this to work?
Like you do for any other lora. I included example prompts.
@asdrabael When i click the create button here, it opens the creator window but it isnt shown anywhere there. I cant select it like other resources. it doesnt show up at all even when i search for it.
@Boobtopia same problem :(
@2busty2868
@Boobtopia
No clue. It's civitais website breaking as it normally does. I recommend generating locally or using a cloud service like runpod so you don't have to experience the issues
I like this lora quite a bit. For me, it does a great job of generating realistic-looking movements. Unfortunately, it also softens the image/video. For instance, a man or a woman who, in my image, has defined abs, will not have those abs in the finished video.
I've tried different settings on lora weights, steps, and cfg, but haven't been able to figure out how to prevent this.
Does anyone have any suggestions? Thanks!
Probably a result of none of my trained people having defined musculature. I went more for a softer body jiggle type motion so that the breasts, ass, and everything else moves. Abs don't really jiggle.
asdrabael Makes sense. Regardless, your lora is the only one of 3-4 I've tried with which I can get consistently decent results. Thanks!
TheCunningLinguist
It's no problem. It's probably because I'm fairly sure the dataset I used for this is much larger than 99% of the lora's. Because of the size this lora took me like a week to train since I did it entirely on my home pc. I've thought about seeing if I could improve it, but the effort doesn't seem to be worth the time since it already works so well in general. The biggest flaw I found was that it's absolutely awful at showing dudes dick's if they withdraw them. I wanted the lora to be able to show the penis withdrawn, and then put back in and it just never worked right.
asdrabael Well, don't feel bad. I'm not sure I've tried a single lora that does that really well. I have a whole bunch of WAN loras, but I end up using only 6-7 on the regular, and yours is one of them.
asdrabael One more question, if you don't mind. Do you have any tips for prompts to more consistently generate thrusting motions, and, specifically, harder and faster thrusting motions. If I generate enough videos, I'll get a couple with really good motion, and sometimes harder/rougher. Do you have any suggestions on good prompts to increase the consistency with that?
My videos are coming in slow motion with this lora, any advice to prevent this?
Did you figure anything out?
@beng2 I combined this lora with another doggy style lora and also added 2.2 wan general NSFW lora i get slow motion from time to time but its less likely now, adding words like fast to the prompt helps too but its not necessary for me now
It seems like almost every example is also in slow mo
do have an example workflow for this?
how do i put this in a high and low noise workflow?
we really need this updated to wan 2.2
Please don't ever stop with these, lol ... this is the first I've tried that actually works and works surprisingly well too. You seem to be good at this.
Hi. Thank you for this lora. It's great. Unfortunately, I have found that the "Penis lora" lora changes bodies a lot and yours alone doesn't. If you can make penises appear more consistently that would be great. I really don't want to have to use that penis lora.
Edit: I may be using a different one because the one I have, even though it's the same size has separate low and high noise files.
Bro, you know why I have always "slow motion" style ? Impossible to have natural movement ? Its amazing LORA thanks for the work.
its because when I trained it, I didn't normalize the fps on the videos being trained to the fps that wan does naturally. some of the videos had 30+ fps so it caused a slow motion effect to happen. it can be fixed with interpolation after you make it but it's just an oversight on my part.