This Is What Happens When You Re-Upload a YouTube Video 1,000 Times

One year ago, someone decided to test the image and sound degradation that occurs when you upload a video to YouTube, then download the YouTube result, and upload it again. He did it 1,000 times, with trippy results.


The process is simple, analogous to the effect of photocopying a photocopy again and again.

Every time you upload the resulting video to YouTube, their video servers will re-encode it again. Encoding the video means that it will be compressed, taking details out of the image and audio, and producing artifacts. When you do this once, the details and the artifacts are not noticeable. That's how compression works: A video gets smaller in size—and therefore easier to transmit through the internet—thanks to the brain's ability to ignore the lack of details. The brain—such a wonderful and forgiving machine—fills in the blanks and problems that the compression program thinks you won't notice.

This is the original YouTube video, which uses a compression standard called H.264:

The problem occurs when you go through this process many times. The artifacts and lack of detail get fed again to the compressor, which takes out even more details and introduces new glitches. The second time you do it, you probably won't notice it. But every time this process is repeated, more will go off.

Here you can see the YouTube video after 56 re-compressions:

And here it is after 474 re-compressions:

The final result is beautiful, albeit psychedelic and strange. Almost impressionistic.


Canzona, the author of this video, says his work is as an homage to Alvin Lucier's I'm Sitting in a Room art project, in which Lucier recorded his voice in a room, then played it back and recorded it again, repeating the process until the echo effect in the round distorted it completely.

When he started the process a year ago, Canzona expected the final result to be a mess of digital video noise, and the audio to be much clearer, but it turned out differently. [Ontologist]




I guess I'm a little surprised the sound degraded so quickly and so completely... far faster than the video did.

I would have thought the algorithm, at some point, would examine the video, see that it's "compressed enough", and just accept it as-is. I mean, you can easily determine the resolution, running time, and file size... can't you use that to make a decision on whether to re-encode a file or not?