I've used the same data set I used for my wan 2.1 I2V and trained a concept lora for Wan 2.2 T2V.
From my testing, I used the lightx2v 4 steps lora with the resolution of 512x512 with 120 Frames. I recommend to use a total of 8 steps.
I used the "/diffusion-pipe:v20"-template from the creator antilopax on runpod.
Trigger word:
thigh sexAn example would be:
A video of a woman having thigh sex with a man. The woman moves her hips back and forth.Description
FAQ
Comments (7)
try this prompt with Headshot or photo smaller than half body to i2v.(stand maybe can change to lying)
"The view is Side viewzooms out to reveal A man stands behind A lower body na-ked woman having thigh se-x . The woman moves her hips back and forth.viewer can see the man's pe-nis sliding back and forth between her thighs.Fixed camera.The woman's face remains on screen from beginning to end."
hope work for wan 2.2 i2v?
It does something. I think it will add motion, but thats prettyy much it. Feel free to test
it works great for i2v
I’m getting nonhuman Pixar type females instead of human women, any suggestions to fix this?
If you tried to use it on civit, the only thing i can tell you is to make sure to use both loras (high and low)
@AbsoluteBussin Doing it offline, I'm not sure what's up. But thanks for making it nonetheless
Details
Files
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Mirrors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
T2V_Thighsex_LOW.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors
Wan2_2_T2V_Thigh_sex_LowNoise-000128.safetensors