Im glad this was posted, can someone explain this deep fake thing to me?
You could watch the Corridor Digital video that was posted, but essentially it's using neural network artificial intelligence. Which is to say it's a form of machine learning, like teaching a person to do a hyper-specific task. Except in this case, it's not a person, it's computer code.
If that sounds scary in a "robots will take over the world" way, it's
not, necessarily. At least not yet. Though these are thinking and learning computers, they are incredibly single-minded and are literally trained to only do one thing. They also, at least so far, have to be trained by people. What this means is that you have to input data (in this case, reference photos from different angles and expressions, both of the source and target) and then when the computer starts generating results from its learning process, a human person must tell the computer which results are better and which ones are worse so that it can improve its accuracy.
And the computer learns and learns and learns, examining and comparing the reference photos, doing hundreds, thousands, maybe even millions of iterations in order to accurately simulate one person's face (Jim Carey) doing the movements of another person's face (Jack Nicholson). That Shining video in the OP could be the result of days worth of learning and rendering.