View Full Version : DeepFake Videos - can't always believe what you see

10-15-2018, 04:07 PM
As reality tumbles deeper into the realm of cyberpunk science fiction, a separate thread on the DeepFake seems warranted.

In a recent video published by researchers to show how the technology works, an actor sits in front of a camera moving his face. The computer then generates the same expressions in real time on an existing video of Barack Obama. When the actor shakes his head, the former president shakes his head as well. When he speaks, Mr. Obama speaks as well.

“This is a big deal,” Hany Farid, computer science professor at Dartmouth College, told The Wall Street Journal. “You can literally put into a person’s mouth anything you want.”

Prof. Christian Theobalt, part of a team working on the technology at the Max-Planck-Institute for Informatics in Germany, said he is motivated by the creative possibilities that it holds for the future.


10-15-2018, 04:09 PM
Fake news sucks, and as those eerily accurate videos of a lip-synced Barack Obama demonstrated last year, it’s soon going to get a hell of a lot worse. As a newly revealed video-manipulation system shows, super-realistic fake videos are improving faster than some of us thought possible.

The SIGGRAPH 2018 computer graphics and design conference is scheduled for August 12 to 16 in Vancouver, British Columbia, but we’re already getting a taste of the jaw-dropping technologies that are set to go on display.

One of these systems, dubbed Deep Video Portraits, shows the dramatic extent to which deepfake videos are improving. The manipulated Obama video from last year, developed at the University of Washington, was pretty cool, but it only involved facial expressions, and it was pretty obviously an imitation. The exercise served as an important proof-of-concept, showcasing the scary potential of deepfakes—highly realistic, computer-generated fake videos. Well, that future, as the new Deep Video Portraits technology shows, is getting here pretty damned fast.


10-15-2018, 04:10 PM
The tool is known as FakeApp, but the videos are usually called “Deepfakes” because that’s the handle used by the original developer on Reddit. You can download the code freely all over the internet, but it’s not easy to set up — you need to configure Nvidia’s CUDA framework to run the FakeApp TensorFlow code, so the app requires a GeForce GPU. The video you want to alter has to be split into individual frames, and you need a large number of high-resolution photos of the face you want to insert. In the case of Obama, there are plenty of photos online that can be used to generate a model.


11-16-2018, 02:28 PM
"This is my very first day at Xinhua News Agency," says a sharply dressed artificial intelligence news anchor. "I look forward to bringing you the brand new news experiences."

As foretold by the scriptwriter sages, many many moons ago.