Disney demonstrates stunning high definition deepfake

Researchers at Disney studios have unveiled a new technique for creating deepfakes. Using deep learning, they can swap faces on videos in high definition and with amazing realism.

What if Disney planned to take advantage of the deepfake to reuse the faces of deceased actors? The methods currently used in cinema to replace faces are very long and expensive to carry out. A new technique, based on artificial intelligence , has just been unveiled by researchers from Disney Research Studios and the Federal Polytechnic of Zurich.

The particularity of this deepfake method is to work in high definition with images of a megapixel. Researchers begin by analyzing faces, recording the movement of certain landmarks. Each face is then normalized in a square of 1,024 x 1,024 pixels , then added to an identity bank. The same method is then applied to the source video. The decoder then makes it possible to place one of the prerecorded faces over that of an actor.

A much faster method for cinema

Researchers use deep learning ( deep learning ) with a progressive drive, combined with a stabilizing system and an algorithm to maintain the brightness and contrast. The results of this new technique are much more realistic than with the previous methods, especially on the big screen.

Although the researchers do not comment on the possible uses, these high definition deepfakes would allow a studio like Disney to revive dead actors in their films. An anonymous actor would play the role, and the studio could then replace his face with that of any celebrity.