uncutWANg
-----------------
An anatomy model for uncircumcised wangs
-----------------
This Lora model was trained for about 40-ish hours on Wan2.1-T2V-14B , it can also be used for I2V and VACE at lower strengths (0.65 - 0.80)
-----------------
Images/Clips posted as examples are unedited and generated from text only
(except for one that shows I2V compatibility)
About:
The Lora provides basic anatomy knowledge for the base model to understand how an uncircumcised penis looks at a soft/flaccid state. No interactions or faces were included in training which should hopefully eliminate unwanted reactions with other loras. The images and videos were masked during training to prevent background bleed, a future update with further masking should fix the remaining quirks.
(it likes muscular body types and a certain chest hair pattern at high strength)
The training material included 40 high resolution neck-to-thigh/floor images from different angles as well as 20-ish videos of the same type either walking, sitting, standing still or lounging.
Generations with a moving subject work well, generations without motion may still produce some movement from the little guy, as training material without someone doing the helicopter or jorkin it was hard to come by.
Interaction loras that work with this model are on the way that will add some needed interactivity.
VACE can also add some interactivity but it's currently limited unless the control video is highly detailed.
Settings:
For T2V, a strength of 1 works well but can be lowered to 0.80 before details get lost.
For I2V, a strength of 0.70 is best to start with (higher strengths can work but ymmv)
Trigger(s):
uncutwang, uncircumcised, uncut, flaccid, penis
All work together or alone, some combinations produce different effects.
For a partially visible head, the video caption that it learned from is:
a nude man from the shoulders down with a brown skin tone large uncutwang penis, uncircumsiced, uncut, flaccid, hairy, shirt, watch
Individual words usually produce fully covered glans, sometimes including "vertical slit at tip" or "drip" will add that detail if it is missing.
Training Files:
Anyone who wants training files can have them, or if you want to know more, send me a message and I'll message back when I can.
Description
1.0
This first version adds basic anatomy knowledge, interaction is not trained so prompting any direct contact with the anatomy will most likely not work. Gravity should interact nicely though.
FAQ
Comments (11)
What a bunch of show offs. I'd let it all hang out too if I was packing that much!
bueno...
My man, thanks for this. I've been trying to wrestle convincing foreskins out of WAN for way too long.
It'd be ever better if this could serve as a "general purpose" uncut shlong generator and work in tandem with BJ/HJ LoRAs, too.
Again, good stuff!
I got it kind of working in I2V by combining it with dreamJob and messing with the strength, the only issue was that it failed to get the size of the head right (too small in this case)
this works for img2video for all big floppy wangs even if they dont have foreskin
not working on wan2.2 t2v
doesn't work on wan2.2
It does work fine for 2.2 T2I as long as you are using 2.1 VAE. Hoping you will work on a dedicated 2.2 version. Your training set is fire.
Is this model "High" or "Low"?
Best lora for Flaccid dongs, that i'm sure about it. at least realistic ones.
Best realistic dong for wan, especially the physics. I only wish you'd add an erect version.