AVATAR model is only for face-on videos like this example. I know some users have been trying to make regular deepfake videos with multiple scenes with AVATAR, but it won't work.
Did anyone ever experiment with Avatar? I've got a few things I wouldn't mind trying out with it but since it's a pretty old build it may not be worth bothering with given the current state of the art: https://niessnerlab.org/projects/thies2019neural.html