Motive is a new way to figure out which training videos teach an AI how to move things realistically, not just how they look.
This paper introduces Knot Forcing, a way to make talking-head videos that look great while being generated live, frame by frame.
This paper teaches a computer to turn one single picture into a moving 3D scene that stays consistent from every camera angle.