4 ms·

Deep fakes typically are putting a face onto an actor (or other footage). Not faking an entire scene with multiple people, that's just called CGI/Animation. S

Deep fakes typically are putting a face onto an actor (or other footage). Not faking an entire scene with multiple people, that's just called CGI/Animation. So you could easily take existing footage and change who it appears to be. But, 1) doing that to frame or unframe someone will be pretty niche 2) the original correct footage might still be findable 3) with access to the video files(which is going to be required in any legal situation), I'm sure there will be (exists for photos) algorytms that detect the editing the deepfake did to video file.

strogonoff 6y ago
Deepfaking doesn’t have to be used on its own. A dedicated entity can shoot entirely new original scenes (unique footage), deepfake actors’ faces to look like the people they’re targeting, edit it all together, and possibly add CG VFX as a final layer to make the resulting video look very real[0]. Of course, knowing all that is not really necessary for someone to shrug off truthful incriminating footage as “fake news”. [0] One caveat is that deepfaking works better with higher quality footage. To produce a complicated scene with altered actor faces while maintaining realistic “phone video” look, it would make sense to film with good lighting and high-quality gear into log or raw format, and then imitate the look in post-production after deepfaking is applied. Thus, one method of detecting such fakes could be by checking for traces of VFX, artificial noise, signature lens properties, signature behavior of phone video recording “magic” (such as noise reduction and stabilization), etc. Enough of producer’s dedication could make that tricky, but IMO it could be easier than applying automated deepfake detection straight up—it’d be buried early enough in post-production workflow, with a lot of noise introduced by subsequent “phone look” VFX.
- nobody9999 6y ago
  >Deepfaking doesn’t have to be used on its own. A dedicated entity can shoot entirely new original scenes (unique footage), deepfake actors’ faces to look like the people they’re targeting, edit it all together, and possibly add CG VFX as a final layer to make the resulting video look very real This was treated (along with the ubiquity of CCTV in London) by the 2019 BBC show The Capture[0]. They glossed over a bunch of technical issues, but got the idea across pretty well. Of course, they made the Americans look like the worse bad guys (no one comes out looking good, even the nominal "victim.") but that's to be expected, given the darkness the show attempts to purvey. It's not great, but it absolutely provides context for the eventual advent of reasonable quality deep-fakes and the potential for abuse of the technology. I'd expect that we'll see lots more of this sort of story telling, especially since any sort of video production (not including raw video footage) is never seen the way things actually are. As such, the "faking" (minus the deep-fakes with faces, etc.) is the key to the exercise. Go watch the taping/filming of any television or movie and that will become immediately clear. [0] https://en.wikipedia.org/wiki/The_Capture_(TV_series) https://en.wikipedia.org/wiki/The_Capture_(TV_series) Edit: Added the missing link.

strogonoff 6y ago

Deepfaking doesn’t have to be used on its own. A dedicated entity can shoot entirely new original scenes (unique footage), deepfake actors’ faces to look like the people they’re targeting, edit it all together, and possibly add CG VFX as a final layer to make the resulting video look very real[0]. Of course, knowing all that is not really necessary for someone to shrug off truthful incriminating footage as “fake news”. [0] One caveat is that deepfaking works better with higher quality footage. To produce a complicated scene with altered actor faces while maintaining realistic “phone video” look, it would make sense to film with good lighting and high-quality gear into log or raw format, and then imitate the look in post-production after deepfaking is applied. Thus, one method of detecting such fakes could be by checking for traces of VFX, artificial noise, signature lens properties, signature behavior of phone video recording “magic” (such as noise reduction and stabilization), etc. Enough of producer’s dedication could make that tricky, but IMO it could be easier than applying automated deepfake detection straight up—it’d be buried early enough in post-production workflow, with a lot of noise introduced by subsequent “phone look” VFX.

nobody9999 6y ago
>Deepfaking doesn’t have to be used on its own. A dedicated entity can shoot entirely new original scenes (unique footage), deepfake actors’ faces to look like the people they’re targeting, edit it all together, and possibly add CG VFX as a final layer to make the resulting video look very real This was treated (along with the ubiquity of CCTV in London) by the 2019 BBC show The Capture[0]. They glossed over a bunch of technical issues, but got the idea across pretty well. Of course, they made the Americans look like the worse bad guys (no one comes out looking good, even the nominal "victim.") but that's to be expected, given the darkness the show attempts to purvey. It's not great, but it absolutely provides context for the eventual advent of reasonable quality deep-fakes and the potential for abuse of the technology. I'd expect that we'll see lots more of this sort of story telling, especially since any sort of video production (not including raw video footage) is never seen the way things actually are. As such, the "faking" (minus the deep-fakes with faces, etc.) is the key to the exercise. Go watch the taping/filming of any television or movie and that will become immediately clear. [0] https://en.wikipedia.org/wiki/The_Capture_(TV_series) https://en.wikipedia.org/wiki/The_Capture_(TV_series) Edit: Added the missing link.