How one deepfake prankster is changing cinema history
“I’m not a coder, just a user,” said Ctrl Shift Face’s creator. “I don’t know the details about exactly how the software works. The workflow works like this: You add source and destination videos, then one neural network will detect and extract faces. Some data cleanup and manual extraction is needed. Next, the software analyzes and learns these faces. This step can sometimes take a few days. The more the network learns, the more detailed the result will be. In the final step, you combine these two and the result is your deepfake. There’s sometimes a bit of post-process needed as well.”
Things could be about to get even more convincing, too. While the attempt to replace Arnie’s audio on Terminator 2 didn’t work as planned, there are efforts afoot in research labs to pioneer the audio equivalent of deepfakes. That means that, in the near future, it’ll be feasible to realistically recreate an actor’s speech using similar training data.
“That’s out of my expertise,” Ctrl Shift Face’s creator said. “But I may collaborate with people that can do these types of AI voices in the future.”
See the full story here: https://www.digitaltrends.com/cool-tech/ctrl-shift-face-deepfake-changing-hollywood-history/