A University of Washington research team has released a new video of Barack Obama in which they superimposed his mouth speaking about the June 12 shooting at the Pulse nightclub in Orlando onto an entirely different recording of his face. The result is a hybrid video that looks and sounds authentic, but is fake, and it’s especially troubling when you consider how this technology could be used to make fake news videos in the future.
How the technology works
The research team used a computer system called a “neural network study” to analyze hours of video footage of Obama speaking to the camera. They then used a computer animation program to realistically recreate Obama’s mouth movements, even nailing details like his lip color, teeth, tongue and oral wrinkles.
After they had an animated recreation of his mouth, they superimposed it onto a video of Obama speaking in a different place altogether — so while the words are his, he didn’t say them in the location shown.
The researchers also took audio recordings of Obama from different points in his political career and made four separate videos of him saying the exact same things in four entirely different locations, wearing four entirely different sets of shirts and ties. His body and facial reactions are real, but his mouth is not.
Here’s a video that shows exactly what they did:
What’s disturbing about this video?
The video above also shows the research team taking Obama’s old speeches, rearranging the words and then creating a video that makes it look as if he’s actually saying the newly arranged words in a single seamless video take. They did the same thing with an Obama impersonator, recording the impersonator’s words and then making it seem as if they’re coming out of the real Obama’s mouth.
While the technology is undoubtedly mind-blowing, the potential for abuse is scary.
Let’s be clear: The technology is not yet advanced enough to create fake videos of just anyone saying anything — thus far, the technology only works well with Obama because there’s so much video footage of him speaking directly into a camera. As a result, his animated mouth is always front-facing and never turns or dramatically changes shape.
However, with advanced technology and enough hours of footage, fake news peddlers could conceivably create videos of politicians or celebrities saying all sorts of incendiary things that were either rearranged from their past speeches or voiced by an impersonator.
This future may not be too far off considering that a Canadian company recently unveiled software that can imitate people’s voices. Right now the recreated impersonations sound a bit robotic, but they may sound more realistic as the technology improves.
Considering the real world impact that fake news can have — remember the Pizzagate gunman? — the potential abuse of this technology make it worth keeping an eye on.