Better Video Chat Through AI Frame Prediction

This is the­o­ret­i­cal as. But say you trained a NN to take the pre­vi­ous 120 frames and pre­dict the next 120 frames of a video, and you could run it 60 times a sec­ond. You could then have a video play­back that played the next frame based on the NN, but then pre­dict­ed the next frame based on what came in. My the­o­ry is that it would be good enough to pre­dict the next frames and essen­tial­ly pro­duce smooth (fake) video from a poten­tial­ly chop­py image. If it could run fast enough.

The cool thing about this is that you would­n’t actu­al­ly be watch­ing a live per­son, you’d be watch­ing the AI pre­dic­tion of a live per­son, pro­ject­ed X frames into the future — this would effec­tive­ly mean there is no lag because the AI is good enough to pre­dict what you’re going to say/act and is always adjust­ing based on what you did say/do. We’re talk­ing sub-sec­ond here, so it’s not pre­dict­ing what you’re going to say, it’s pre­dict­ing the changes in tone and pitch and where your face can get to in the next sec­ond. That part is total­ly pos­si­ble. The part I think would be impos­si­ble is that you could­n’t do this fast enough and dis­play the result. Maybe quan­tum computers.

This entry was posted in AI, Software. Bookmark the permalink.

2 Responses to Better Video Chat Through AI Frame Prediction

  1. Chris says:

    The ver­sion of this I liked was to reduce your face to vec­tors using OpenCV, then deep­fake your own face back onto the vec­tor skele­ton at the oth­er end. You’d need a chunk of data at the start to get things mov­ing, but then absolute­ly min­i­mal data to send the vec­tor skele­ton at a but­tery smoothe 120fps

Leave a Reply

Your email address will not be published. Required fields are marked *