Watch What Happens When an Algorithm Tries to Predict the Next Frame 100,000 Times in a Row

Like an athlete trying to push their body to its extreme limits, artist Damien Henry was curious what would happen if you asked a predictive algorithm to calculate the next frame of video in a sequence, again and again and again, over 100,000 times.

The results are a fascinating music video that looks like overly compressed digital video filmed through the window of a moving car. The experiment was not unlike repeatedly making a photocopy of a photocopy, given that Henry fed the software just a single image to start with. But the machine managed to generate over 56 minutes of unique video that somehow remains fascinating to watch all the way through.


Predictive algorithms like this, powered by machine learning, are useful for more than just mind-melting art projects—Google has already shown how they can be used to generate moving video from still images. Eventually, they might allow anyone to easily create their own virtual worlds with VR, or let someone working on a laptop create the next blockbuster movie without a Hollywood-sized budget. For now, though, if you’re looking for something to fall asleep to, this should do the trick.

[YouTube via prosthetic knowledge]

Share This Story

Get our `newsletter`


What was the first image fed in?

Because that is definitely grass, trees and blue sky going past, and that doesnt appear until partway through...

If the original image didnt have trees and sky in