This makes the subject's face and hands move like they are a bit wackadoodle, aka cuckoo for coco puffs, aka one clown short of a circus.
There's no low noise model because this high noise lora is only for movement. It was trained for i2v, and the examples are all i2v. But it should also work for t2v. Note that it might only work for closeup portraits.
I was trying to train something else, but had to cut training very early, and this was the result. Why would you want this? IDK, but here you go.
