Facebook AI scans videos of real people to turns them into game characters

To train Pose2Pose system, the researchers used real videos of a tennis player, a person swinging sword indoors, and a person walking.

Facebook new AI experiment on gaming. (REUTERS)

Facebook has developed an artificial intelligence system that is capable of making your favourite friends from real-world turn into playable characters of a game.

The system, called Pose2Pose is a framework that uses the references from the real-world videos to build novel image sequences for a virtual character that can be controlled using joystick or keyboard, Venture Beat reports.

To train the system, the researchers used real videos of a tennis player, a person swinging sword indoors, and a person walking. Using the reference, the system was able to successfully produce dynamic elements such as varied camera angle and character clothing.

The researchers believe the Pose2Pose can pave way for a new generation of video games with realistic graphics.