News for hackers. Stuff that matters.
Re:Show HN: Create a Deep Fake Video from Any Image by abeppu on Thursday June 04, 2020 @08:55PM (#23423889)
There's a huge and qualitative difference between the example video and the 'recent videos' below.

The example video shows the face being stretched mostly around the mouth, and without extreme or sudden changes. Sure, it looks fake, but it seems coherent. It's almost the kind of thing a human would make in a comedy setting. But in the generated videos below, I see a lot of the issues we often get with generative models -- eyes or mouths flickering in and out of existence, a disappearing neck, parts of the background getting mixed with the face, etc. Sort of nightmare-ish, but not at the level of deep-dream dog-worms from 2014 or whenever.

Were these even from the same method? Or is there some about the user-submitted photos that meaningfully differs from those used during training?

Re:Show HN: Create a Deep Fake Video from Any Image by soheil on Thursday June 04, 2020 @10:07PM (#23424294)
The instruction wasn't clear, the video and the image both need to have a person's head for this to work.
Re:Show HN: Create a Deep Fake Video from Any Image by jcims on Thursday June 04, 2020 @10:39PM (#23424465)
Is this related to avatarify at all? I tried that a few times and the result was very similar to what is in the recent videos section.