Lifelike Audio-Driven Talking Faces Generated in Real Time

[Ken_N]

From a single photo, and a short sample of audio, very realistic video can be generated and streamed in real time.

https://www.microsoft.com/en-us/rese...roject/vasa-1/


“… Our method is capable of not only producing precious lip-audio synchronization, but also generating a large spectrum of expressive facial nuances and natural head motions. It can handle arbitary-length audio and stably output seamless talking face videos...”

Looks like for now Microsoft does not plan to release this, they are keeping it for themselves.

[james112]

Wow, I would have been completely fooled these are not real people. These are very realistic. Has significant implications. Why hire an actor for a tv ad, or even for a TV show? Or even for a movie?


It could also be used to fake what someone said on video, when they never did. Which should not be used for say in a court of law. All you need is someone's photo, and it can generate life-like speech. No wonder they are not releasing it.