Trained with https://github.com/PGCRT/musubi-tuner_Wan2.2_GUI on a RTX4080. I used 12 clips, ranging from 3 seconds to 6 seconds, for the dataset.
Both HIGH and LOW models are needed for it to work well. All the example videos were generated with Lightx2v LoRA. Workflows are included in the sample videos.
You might get, what I call IDS (Invisible Dick Syndrome) in some angles. NSFW Fix LoRA mitigates the issue, but I'm not sure if it completely fixes it. Just try another seed or another angle through prompting when that happens.
Recommended Settings:
Steps: 8 (4/4)
CFG: 1
Shift: 8
Scheduler: Euler, Beta57
Trigger word:
pr0ne, a man is penetrating a woman from behind.You can describe the details and the scenery as you like.
Prompt examples:
pr0ne, A nude, muscular man with light skin and brown hair is penetrating a woman with tan skin and long black hair from behind. She appears to be experiencing pleasure. The scene takes place on a tropical beach. medium close-up shot.pr0ne, a muscular man with a well-defined physique is penetrating a pale-skinned, red-haired white woman from behind. The woman is lying on a red and white striped towel, partially nude, with large breasts prominently visible. Her expression is one of surprise or pleasure. The background shows a wooden deck with a rocky, desert-like landscape. The lighting is natural, highlighting the subjects' skin textures.pr0ne, a naked man is penetrating a nude blonde woman with a curvy figure and medium-sized breasts from behind. She is on all fours on a bed with bright blue sheets, eyes closed, and mouth slightly open in pleasure. The camera angle is medium close-up, above eye-level, slightly from top. The background shows a dimly lit room with patterned wallpaper and a partially visible bathroom. The image is well-lit, with natural light highlighting the subjects.Buzzes are appreciated. I didn't put it behind a paywall so you can start playing with it right away.
Enjoy!
Description
Initial release
FAQ
Comments (19)
Awesome work, beat me to it, saved me some bucks thank you haha.
Haha glad to help. 🫡
I've been waiting for this pose for a long time. Thank you very much!
que pose mas rica! gracias por este lora. lets play!
wow, nice output
Is there an ITV version?
No there isn't but give t2v one a try. In my (very limited) tests, it worked well when the image is a similar pose/angle to the ones in the sample videos. Adding the NSFW Fix lora also helps in this case.
I am having a issue that pose and visual character quality are looking great but it is missing the male gen very often in these upper side side angles as if it would be cut off and it is like this in in various workflows which you mention in your description.
What i tried:
The mentioned NSFW LoRA but does not resolve this on my side.
I set a weight of 1 for both H&L and generated in in the example workflows set resolution.
And added the trigger and the example prompts.
Different supportive LoRA's like lightning 4 Steps and the default default workflow embedded 4 Steps T2V LoRA.
Using the offered PNG embedded default workflows like drag drop generate.
While it is working in the video it does not using the same workflow metadata then.
I sure checked for all LoRA are added new since your paths are different.
Are there any more hints beside the already given or other peoples experiences?
The issue comes from the WAN model not being trained on the male genitalia. (and probably amplified by the fact that some of the videos in the dataset I used to train this lora have the genitalia hidden behind the woman's butt or man's arm)
As I mentioned in the decription, NSFW Fix lora sometimes helps to mitigate the issue but it's not a 100% solution.
You can try adding another lora at low strength (0.3~0.5), in which the male genitalia is more prominently visible. (not tested but this one could help https://civitai.com/models/1811313/wan-dr34ml4y-all-in-one-nsfw)
No dicks. Needs a v2
You might be referring to as the OP calls it, an IDS (Invisible Dick Syndrome) in some angles. As per OP, NSFW Fix LoRA mitigates the issue
@honryindian Didn't do anything to fix it on mine. It needs to be retrained with close up videos.
Look closely at almost all the examples. No dicks. It's a good v1, but not good enough.
@xicen29527524dfdfd You can try the other Prone Bone LoRA, looks really good!
@honryindian Oh shit. @playtime_ai made one? Say less! I'm on it.
I don't know if this is intentional but this LORA is usable with WAN V2.1 I2V with the online generator. I have tried twice now to generate with it using 2.1 and both attempts have produced blurry pixilated results.
Not sure if this is something that can be corrected server side or client side or even if this LORA was intended to be used as I2V, but at least I figured I would make an attempt to make someone aware of the issue.
I'm not sure why it shows up as usable for Wan2.1, it's certainly not intentional.
Never used the online generator so I don't know if it's a setting I need to change on my end but I published it for Wan2.2 T2V specifically, which is what it was trained on. It can work for Wan2.2 I2V as well, but not for Wan2.1.
@schwadorf is there a way to make it available for 2.2 I2V on site generation? I'll give it a try and see if I can get some decent results.
can you make a POV version? (from behind)
Weirdly, the best side-view doggy lora I've found.
Details
Files
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Mirrors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise.safetensors
schwadorf_pronebone-wan2.2-t2v-high-14b.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Wan2.2_Pronebone_t2v_HighNoise-000024.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.