Full nelson sex position for Wan I2V.
All the videos in the showcase were created using variations of this basic prompt:
"fullnelson. A video of a man and woman having anal sex in the full nelson position. The woman is on top of the man and is being penetrated by the man's penis in her ass. The man is thrusting his penis hard and fast into the woman's ass repeatedly causing her feet to to bounce up and down in rhythm to the man's thrusts."
The LoRA works for both anal and vaginal sex, just change your prompt to reflect which one.
Works well for the complete full nelson position with the hands behind the neck, or with just the man holding the legs back. In my testing I found it can work decent for the reverse cowgirl position too.
It was trained on as diverse of angles of the full nelson position as I could find, it should work on any angle from front facing to side profile view. It was trained on 28 videos for about 56 hours on the 480p 14B I2V model using two 4090s and diffusion-pipe.
Description
FAQ
Comments (28)
Are the gens all gonna have ginormous juggs?!
No, I just like ginormous juggs so all my source images I posted on the showcase had them. In my testing all breasts sizes worked with it. It can do anime style videos too. It was trained on real life videos with the women having all different sizes of breasts.
Good question, not everyone is into the enormous, cartoonish 'balloon tits' thing. Hopefully the LoRA is flexible enough to mid to normal sized large breasts, too.
@Agkistrodon In my testing it worked with all breast sizes! If your source image has small/normal sized breasts it won’t change them.
@lazerblazer Good to know. It makes sense that the model would follow the size of the breasts on the source image.
You are a legend!!! can't wait to test it out, will let you know the result!
Hope you enjoy and it works well for you!
@lazerblazer Thanks for the Lora again! it works kinda well! As the other review also point out, it seems that the deeper penetration seems to be missing. Tested with photo realistic or anime style and overall the lora managed to capture the fullnelson well! So far the prompt mostly create relatively subtle shaky insertion. But it could be my prompt was not clear enough.. will try more. anyhow, great Lora!
That's a great lora! Thank you! Could you also make a deep penetration version? All in and all out? That would be awesome!
Thanks!
Possibly, I would definitely like to. This was trained on videos that only had deep penetration and this was the result. I’m not sure why it didn’t translate that well. I will have to think on how I can improve my dataset to make it better for the next version.
@lazerblazer good luck!
Hi everyone,
I'm planning to train a character LoRA for Wan2.1 (specifically the 14B model) using the Diffusion-Pipe framework on WSL. My goal is to generate short video clips featuring a specific person.
I have both still images and short video clips (e.g., 3 seconds @ 16 FPS) of the person. I was wondering if Diffusion-Pipe supports training a LoRA using a mixed dataset containing both image (type = "image") and video (type = "video") subsets within a single training run, referencing the same concept folders?
For example, could I configure my dataset.toml like this (simplified):
[[datasets]]
type = "image"
image_dir = "/path/to/my_data/"
subset_dir = "20_my_person/"
# ... other image params
[[datasets]]
type = "video"
video_dir = "/path/to/my_data/"
subset_dir = "20_my_person/"
# ... other video params
Use code with caution.
Toml
Or is it necessary to train exclusively on images or exclusively on videos when using a T2V base model for this kind of LoRA? Does Diffusion-Pipe handle merging or alternating between these different data types smoothly during training?
Any insights or examples of how to configure the .toml for such a mixed dataset would be greatly appreciated!
Thanks!
It supports both images and videos. I’m not sure about the .toml file in from the top of my head, but it’s similar to that. I think the diffusion pipe GitHub has an example that has both
Great LORA, and great description of its use. Any way we could help train it for deeper penetration in future versions?
I love this lora.
So, It's shameless question, but do you have plans to make a mating press?
I'm glad you like it!
I don't have plans for a mating press lora, but that is not a bad idea... I wonder if any of the existing missionary Loras can do the mating press position though? Have you tried by chance?
@lazerblazer I used i2v and tried several Lora with similar poses, but the results weren’t very good. In particular, the missionary position tends to move sideways rather than up and down. Even the POV versions didn’t make much of a difference.Testicles often get turned into a pussy, and 'penis’ ends up looking like the longest snake I’ve ever seen my life. The results were actually worse than when I didn’t use lora. :S
@swan good to know. Not 100% sure if I will make it yet, but I am looking into it and interested.
@lazerblazer Oh, I’m glad you’re interested. Hopefully, there will be some good news.
I feel like it needs more training unless am doing it wrong. The image just shakes around with guys dick is in there. I was fully expecting the image to go balls deep in there
The model is capable of creating the motion as shown in the showcase. If you can't get that, you are doing it wrong.
@lazerblazer model doesnt go balls deep , not in any of the videos from users or yours (its free so it doesnt matter to complain)
@Animeboy007 Yeah it doesn't go all the way in unfortunately. Does more than shake tho, you can get some thrusts.
@lazerblazer If only I had a better set up I wouldn't mind helping train.
i was just training my hun lora that works perefctly on framepack, and im getting what @Devilday666 says, little motion and the penis doesnt even want to go inside,, thats with my own dataset from my hunyuan lora.. so.. something is going on? @lazerblazer i will test yours now..
@lazerblazer getting the same results with your lora, and with mine.. idk wtf is going on.. is quite frustrating to be training for so many hours to see this.. you say is the issue is the workflow?
@Devilday666 i dont think more training will fix it.. i re-trained the same lora twice and now im at epoc 116 epoc, and still no penis ovement.. is like is fused on the woman.. im training with 480p model, but im starting to thing thats is not the issue here..
seems like we need to crank up the CFG, since almost every new workflow with speed loras uses 1.0
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.