AMPLIFY: Actionless Motion Priors for Robot Learning from Videos (Collins et al.)