Making you wish you had three hands!
Works for WAN2.2 a14b t2v and i2v.
All example videos generated using both high & low lora at strength 1.0
When using V2, prompt for specific breast sizes using she has three {small|medium|large|huge} breasts and prompt for bare breasts with she has three exposed {small|medium|large|huge} breasts
All showcase videos have workflow embedded. (Use the download button to preserve it, if you Right-click/save-as the workflow data gets stripped by the browser)
About V2.1
V2.1 should have less of a bias towards cropping out faces and work better when using WAN2.2 without lightx2v.
Mostly identical training images and videos, but redid the captions and tweaked some of the training settings.
Changes:
Captions now specifically include cropped out faces (which appears to have reduced its bias towards defaulting to cropped faces)
epochs: 100 → 50
Low-noise
lr: 5e-5 → 1e-4
max_t: 0.875 → 0.6
High-noise
min_t: 0.875 → 0.5
timestep_sample_method: logit_normal → uniform
Description
This version should have less of a bias towards cropping out faces and work better when using WAN2.2 without lightx2v.
Mostly identical training images and videos, but redid the captions and tweaked some of the training settings.
Changes:
Captions now specifically include cropped out faces (which appears to have reduced its bias towards defaulting to cropped faces)
epochs: 100 → 50
Low-noise
lr: 5e-5 → 1e-4
max_t: 0.875 → 0.6
High-noise
min_t: 0.875 → 0.5
timestep_sample_method: logit_normal → uniform