It just kinda makes no sense to me. How can you improve the framerate by predicting how the next frame should be rendered while reducing the overhead and not increasing it more than what it already takes to render the scene normally? Like even the simplistic concept of it sounds like pure magic. And yet… It’s real.
Works the same way as any other software optimization: lower quality computation as a shortcut.
The predicted frames don’t use the same full stack of data that a true frame uses to render, they just use the previous frames data and the motion vectors. The rest is a very efficient nueral-network guessing algorithm based on those two pieces of data instead of the full shader stack.