Over the past few years, we’ve seen the rapid rise of “deepfake” technology that uses machine learning to analyze footage of real people — and then churn out convincing video of them doing things they never did or saying things they never said.
Now, researchers are pushing the tech even further — with an algorithm so powerful that it can turn a single frame into a full-motion movie, reminiscent of photographs and paintings in the “Harry Potter” series.
The algorithm, described in a new paper on the preprint server arXiv, can learn to create new video of a person by examining just a few frames of their face — or even just a single shot, though the results aren’t quite as crisp.
In a video by algorithm creator Egor Zakharov, the system whips up convincing video of celebrities, including Justin Bieber, Britney Spears and Neil Patrick Harris, and turns still frames of historical figures Marilyn Monroe, Fyodor Dostoevsky, and even Leonardo Da Vinci’s iconic Mona Lisa into lifelike videos.
The results are extraordinary — and slightly terrifying.
READ MORE: Few-Shot Adversarial Learning of Realistic Neural Talking Head Models [arXiv]
More on deepfakes: These Full-Body Deepfakes are Like Nothing We’ve Ever Seen