> 1) Don't video deepfakes require lots of high-quality input video (which is why they were often made of Obama)? Where would an attacker get this for some rand

> 1) Don't video deepfakes require lots of high-quality input video (which is why they were often made of Obama)? Where would an attacker get this for some rando? I imagine that at some point, or even now, we can use transfer learning for deep fakes and just train existing models on a limited data set for "good enough" deep fakes.