Researchers have introduced a neural network that can instantly change faces. At the same time, she records video in high quality, so it can be used in television.
A new article published by Disney Research in collaboration with ETH Zurich describes a fully automated neural network that can change faces in photos and videos. This is the first method that can change faces in real-time and record video in high quality. AI can be used in film and television, and the quality will be so high that “viewers will not be able to distinguish a real person from an edited one”.
Researchers plan to use this technology to replace the actor’s face when he needs to play an older person. They also suggest that this technology may be useful in cases where a stuntman plays part of the scenes for the actor.
Researchers noted that the new method is unique compared to others, including the fact that any person used on the set can be replaced by any recorded person. However, in most cases, scientists did not notice the “sticks” that can be seen on many other deepfake videos – the actor looked the same as in other scenes.
TechCrunch notes that in some cases, AI does not work well – this is especially noticeable when poor-quality data is supplied to the algorithm. However, in comparison with analogs, it gives a good result several times more often. “Most importantly, the algorithm works with a much higher resolution – this is a key factor for the entertainment industry,” the publication notes.