Guy uploads, downloads, then re-uploads the same video to YouTube 1,000 times | Hacker News Reader

Guy uploads, downloads, then re-uploads the same video to YouTube 1,000 times

slate.me

209 pointsjazzychad133 comments

Was the aspect of using YouTube's servers so critical to the artistic quality of this project that he couldn't just substitute a shell script that encodes the video 1000 times over? Would've taken a heck of a lot less time than the year he spent.

The benefit is that all 1000 videos are available on YouTube - you can check out the quality at any step along the way.

Maybe I'm overthinking it, but couldn't he have written a shell script that did the uploading and dowloading to Youtube? How would we know if he did?

Despite what you might think, the majority of the planet has no clue what a shell script is.

Savages.

http://en.wikipedia.org/wiki/Rational_ignorance

Good point, but it's possible that he enjoyed the process somewhat. Recording 'I am sitting in a room' probably became a little tedious for Lucier as well, in all the time he spent from conceptualization until completion.

And even if he did not enjoy it at all, patio11 makes another good point about it being a better background story for the piece.

Yes, you are probably right. My comment was aimed at the snobbery over the unwashed masses.

If you can't see the humor in feigning elitism over shell scripting (shell scripting!) I really can't help you.

Yes, that interpretation is more charitable. I guess I am too afraid of Unix fanboyism, because I tend to fall into that trap myself, when I am not fighting it.

Perhaps, in addition to the artistic aspect of it, he wanted to educate people about what actually happens to video when it is sent to youtube. This is much more relatable than saying, "I wrote this shell script..." Also, having the log of all 1000 versions on youtube is interesting as well.

The other thing about art is that time is usually not a large motivating factor...

The thing is, I'm not convinced that it's youtube's fault; apparently he converted it to mp4 on his own computer after the download from youtube, before re-uploading it. I wonder if he'd have had the same effect if he'd uploaded the same format he ripped from youtube.

Nobody is blaming youtube, video compression or anything of the such. This is to expected behaviour. And no normal person should be able to expect, or want to upload and download a video 1000 times, or even more than 10.

He does say "and the mp4 codec" in his speech, so that's acknowledged at least.

There are clear reasons why not everyone is an artist, patience and endurance is are a huge part of it.

That's also the same reason many people look down on (modern) artists. The use patience to do something everyone else would do in a more efficient way.

Don't mistake patience and endurance for skill.

They're close enough for me.

1% inspiration, 99% persipration.

“If Edison had a needle to find in a haystack, he would proceed at once with the diligence of the bee to examine straw after straw until he found the object of his search.

I was a sorry witness of such doings, knowing that a little theory and calculation would have saved him ninety per cent of his labor." - Nikola Tesla

I'm sure Edison was crying all the way to the bank over the labour he expended.

Oh, Nikola -- missed your chance for an extra jibe in there. "...a little theory and calculation" -- and a big electromagnet, assuming a ferric needle.

I'd say they're highly correlated. How else do you obtain skill?

He proved that point at #100, I guess. And you really think this as an art?

Not "an art", but certainly "art".

Sometimes the artistic process is just as important as the final product.

A lot of modern art is sold as a story, and for whatever reason the kind of folks who matter think that "did something by hand a thousand times" is a much better story than "wrote a Perl script."

Art isn't the only thing that has stories, incidentally. Tomatoes have stories these days. (OK, the kind of tomatoes rich people eat have stories. Only poor people eat tomatoes that have no stories.)

i think products targeted at low income demographics are very much sold with a story. a kind of utilitarian "sensibleness" which facilitates contempt for more expensive products.

why would advertisers neglect to exploit human psychology at all socioeconomic levels?

I believe you're right... a hugely successful ad for the Tang powdered drink in South America showed a spoilt rich kid asking his butler for Tang after they were stranded with no oranges in sight. (One of the messages I take from the ad was that the rich kid was missing on what the butler knew all along)

Of course it is universal; what is advertising but giving products a compelling narrative?

No, I'm pretty tomatoes are just sold as tomatoes. Produce is produce until you're trying to justify charging a multiplier of the base price.

Not "just sold as tomatoes", but "sold as just tomatoes". See the Sainsbury's Basics http://www.google.co.uk/images?q=sainsburys+basics brand - the story is that this is as basic as it can be, no cruft, implying better value, lower price.

That's one of the common criticisms of modern art, that it is largely about justification of the work.

After I went to Art Basel last year and happened on a display which was, quite literally, a few kilograms of nutella dumped on the floor, I was hard pressed to disagree...

Are you just making the blind assumption that he's a programmer?

One year means he could have spent 364 days learning how to program and do all the 1000 uploads on the final day of the year...

I agree, but...

One day for 1000 iterations means 86.4 seconds to upload to youtube, wait for encoding to finish, then download it.

Youtube is slower than that. :-)

Better give it a week. Maybe a month.

He could work on something else during that month. Automation!

Yes of course, but he can't take 364 days to learn to program.

Or one (maybe a few, I don't know what the going rate for something like this) day working for minimum wage, a few hours to find a freelance programmer, and paying the programmer to write the upload script. Then, spend ~360 days doing whatever.

Something tells me this project was not about efficiency.

I make maple syrup by hand. It requires a huge amount of physical labor, as well as a fairly expensive evaporator and as well as a small building devoted to that single purpose.

The product tastes the same as the stuff I can buy from my neighbor's maple stand for a fraction of the cost and effort. It's wholly irrational.

And I'll do it again next year.

Sometimes the joy comes from doing it the hard way.

Yeah, I immediately thought of this t-shirt when reading your comment :) http://www.thinkgeek.com/tshirts-apparel/unisex/frustrations...

Yes it was. The guy wanted to encode the videos using Youtube's _unknown_ encoding process. Encoding using ffmpeg with its default settings isn't the same. (We can now extrapolate Youtube's encoding logic rather than make blind assumptions, which would be the case if the latter were done instead.)

I'm pretty sure they use (perhaps a custom version of) x264 to encode video.

Because "YouTube" has a meaning for regular people, while "x264" doesn't.

A lot of people I speak to associate x264 with pirated film downloads.

There is a rather large spiritual aspect to mankind's creation of art.

If you look at the history of art, in the western world and elsewhere, it is strongly tied to some sort of religion.

Religious practice typically implements some sort of repetitive behavior or meditation, where the goal is not necessarily to get something done in the physical world.

If you run the process through some sort of reductionist approach, you wouldn't have art to begin with. Why even make a piece of art? What's the point?

While it is interesting to view the creation of art through the lens of an engineer, any attempt to criticize the work through only this lens is not going to stand up very well.

What's the difference between an artist and a Perl programmer? 364 days and 23 hours.

And the Perl programmer has disposable income and a girlfriend.

Well, let's not go overboard here...

Wait, in what part of the world is a programmer more likely than an artist to have a girlfriend?

Ones where girls like the guy to buy them food?

The artist hit the front page.

So did a rumour about .NET code running on the iPhone.

I get and appreciate that joke, but there sure is a smug sense of superiority from the programmers in this thread.

You should see the shit us Perl programmers take from the Ruby guys.

#1000 looks like an alien from another dimension trying to communicate with us.

It's the modern version of photocopying something tons of times..

Certainly a good education about lossy compression for the non-technical public.

The HN community could have saved a lot of time writing a shell script to post a snide comment about writing a shell script from every account instead of forcing everyone in the thread to do it individually.

I think that, simply, this sort of art is quite different from what large parts of the HN crowd would consider artistic/beautiful.

(personally: doesn't do anything for me as art)

The "Angry Video Game Nerd" guy did a similar thing with two VHS tapes, copying a video back and forth over and over:

http://www.cinemassacre.com/2010/05/26/vhs-generation-loss/

so: prior art?!

It's his take on Alvin Lucier's famous "I Am Sitting in a Room":

http://www.youtube.com/watch?v=Jfssj80oNuM

The original recording is linked from the Mashable article.

As was alluded to in the comments there, most of the loss was probably due to stretching and demagnetization of the two tapes. Generational loss would most likely consist of horizontal scanline offset (due to time base skew) and horizontal blurring, though eventually the time may get off far enough to cause some of the other artifacts seen in that video.

I did something like this - I took an image and rotated it a couple hundred times in Windows Explorer. It gradually degraded into a blurry blob.

Which is pathetic, since it's possible to do a lossless (I mean not incrementally lossy) rotation of JPG's.

MS Windows (XP IIRC) used to warn you if you used the built in image rotation function for JPEGs that it was lossy. I've a feeling it was fixed in SP2 but I've not used MS Win for years so am not really sure.

Win Vista & 7 don't have this problem (just checked).

I did this in 1997 or 1998.

Sometimes this effect of multiple lossy compressions is very apparent on certain... ahem... adult sites, which steal content from other adult sites. When the content has been decompressed and recompressed enough times certain artifacts start to appear, as well as other issues like aspect ratio problems.

Does this mean YouTube's compression settings don't target a specific bit-rate but instead aim to make it "no larger than X and at least y% smaller" than what was uploaded?

I'm surprised it's not just a pass-through at a certain point (if nothing else, that'd save CPU cycles).

http://news.ycombinator.com/item?id=1402382

Apparently it was himself who re-encoded the video, not youtube. Dont know where jemfinch got that info though.

And youtube serves content in different resolutions (SD vs HD) and formats (flash vs ipod). It's possible his choice of either of these had a negative effect on the quality. It's also possible it all happened during his on conversion on his local machine.

I wonder what results he would have gotten if he pulled the h264 copy down that is served to apple devices and uploaded it without modifying it.

The process reminds me of this post: http://hackerfactor.com/blog/index.php?/archives/355-How-I-M...

Great post, I wonder if the trick that photoshop is pulling is really beneficial or worse in the long term.

What's really cool about this is that his stated purpose ("to remove all human characteristics from both my voice and my image") is fulfilled by his voice and the individual frames - but the way the shapes move is still very human. A lot more human than CGI animation is still doing, actually.

That's neat.

Now he needs to splice them all into a single video showing the gradual degradation over time.

My speakers were on full blast - the 1,000th uploaded one scared the crap out of me! :O

I'm surprised it continued to change after detail was lost. Seems like there would be some measurable property of a codec/decode process that could be quantified - the entropy? It would be a desirable property to be 'idem-potent', e.g. after one cycle the result settles.

> codec/decode

I think that you meant: 'code/decode.' 'codec' is a abbreviation for 'code/decode.'

I can understand the video getting messed up, but what happened to the audio? From what I understand re-encoding an mp3 doesn't do much - all the waveforms that were going to be thrown out, were already the first time. On the second pass nothing much changes.

Or did he rip the video using speakers and mic, rather than downloading the digital data from youtube?

I have a feeling the errors came during the ripping process, not the encoding process.

re-encoding an mp3 doesn't do much

Sure it does, typically. It’s no different from video in this way: ideally, all of the image features that would get thrown out were thrown out the first time, but the codec isn’t that smart.

Most implementations of lossy compression algorithms don’t have stability under re-encoding as a goal. (A few do – some image editors are smart about passing through unchanged JPEG blocks, for example, and I think mp3wrap will join MP3s cleanly.)

I’m too lazy to look into exactly how he did this, but it’s not impossible that the audio artifacts came from pure MP3^n re-encoding. It’s a case that most compressors ignore.

Why hypothesize? Create a directory. Copy your favorite MP3 into it as both "source.mp3" and "orig.mp3". (orig.mp3 is just for convenient comparison later.) Then:

    for i in {1..100}; do lame source.mp3 dest.mp3; mv -f dest.mp3 source.mp3; done

Then listen to the resulting "source.mp3" at the end. Send whatever params to lame you want in that command line, though the defaults strike me as pretty good for this test.

BTW, when people point out that scripting this is easy, they're not kidding. We're talking shell one-liner here, forget even hitting perl.

I'm on about iteration 7 as I write this and it's definitely breaking down.

Kudos for actually checking, but of course this only proves it for lame with certain flags. There are several popular compressors and many many options. As someone mentioned, YouTube doesn’t say what it uses, but it may be possible to figure it out.

Learning how to do for loops in shell was a big Unix productivity breakthrough for me. I just wish the syntax were cleaner – silly things like quoting rules and not using $ on variables on the LHS keep tripping me up. I’m probably wasting my own time in the long run by not getting good with something like http://www.scsh.net/ .

So I tried it. The biggest problem is that it gets quieter each pass. Not sure why. I tried adding --noreplaygain but it didn't help.

The quality seemed to preserve OK, but as the signal get quieter and quieter the noise started becoming more obvious.

I got silence after about 400 generations. Maybe throw a gain balancing step in there somewhere?

http://www.reddit.com/r/technology/comments/86yiq/hear_what_...

A reddit post on that exact experiment. Sadly, the file is down.

As a side note, I had a post here that I deleted about this, worded something like "Yes, they used that song". I went back to read the topic, and found I'd made nearly the exact same comment on the story over a year ago. I guess I'm a tad predictable.

Edit: Aha! A megaupload link was posted that still works. http://www.megaupload.com/?d=ALZY0LCD

Also related, the same experiment but for .jpg: http://www.reddit.com/r/technology/comments/86v2z/see_what_h...

That file only increased my questions on what happened to the audio on this vid. The song is quite recognizable. So why did the youtube one turn into garble?

And the jpg one is just incorrect. He kept increasing the compression ratio, which proves nothing at all. For a more accurate view see http://hackerfactor.com/blog/index.php?/archives/355-How-I-M... which shows that after a certain point the jpeg stops changing. (Which is what I expected the mp3 to do, but I guess it doesn't.)

The mp3 is probably using a different audio codec than the youtube video/mp4.

Here's the inspiration: http://en.wikipedia.org/wiki/I_Am_Sitting_in_a_Room

I wasn't familiar with the audio piece that inspired this so I downloaded the mp3 from http://www.ubu.com/sound/lucier.html . It turned out to be a nice background soundtrack for programming today.

Music with words usually distracts me when I'm working, but this didn't.

It's really interesting to note the differences in our perceptions of the final audio versus the final video.

The video, while severely degraded, is obviously still a human figure moving around.

The audio, on the other hand, is mangled completely beyond recognition.

Next time the topic of bandwidth caps comes up remember this guy. I don't want to knock his project but I think it illustrates a point that unlimited bandwidth and access to services tends to make people less concerned about using bandwidth efficiently.

He used hundreds of hours of his time. What makes you think he would've refrained from paying a few extra dollars for the bandwidth?

Nothing. I'm sure he probably would have if forced to. I know from my own (bad) habits I often waste bandwidth just because I can. Common example of this is having a file downloaded to my laptop sitting in my backpack but instead of taking it out, waking it up, mounting the share and copying it I'll just re-download it from the Internet instead because I'm lazy and there's really no downside for me.

I'm too lazy to look this up, but did the video get smaller or larger with the iteration number? This is more interesting to me than the look

Time: yes File size: yes

Not sure why the time dropped.

I'm not a video expert, but I did write some software for muxing and demuxing for a company I worked for some time ago. The major problem I had was that video and audio are stored at different "frame rates" (almost in a literal sense). A frame of encoded audio is generally shorter (in seconds of playback) than a single video frame is displayed on screen for. So, if you were cutting and splicing video, you'd end having to decode the audio, break apart the first and last frame's worth of it, re-encode it, and then offset the timing of the audio from the video. This is very obnoxious to get perfect.

I've got no idea if that's what impacted this guy's video though, but I'm sure it's far from the only weird thing involved in video compression.

He mentions this on the comments for the 1000th video, for some reason youtube would sometimes remove the last frame.

Also he had trouble with youtube adding a small offset to the audio, to the point that he had to start editing the videos to re-sync the audio.

Wow. It's like a lesson in wavelet transforms for the layman. But without ever using the word 'wavelet.' Cool.

He probably could have saved a lot of time with a shell script :)

I bet Christopher (prawn protagonist of District 9) might be able to grok the bizarre utterances in the 1000th video... but I can't.

I'm sure Youtube loved getting spammed with a 1000 very similar videos

Probably not as much as they loved an interesting story about YouTube on prominent websites.

come on now, it's youtube, they're the second most used search engine in the world. They don't really need good press anymore

In that same year, how many much less interesting videos do you think were uploaded? I doubt they noticed.

Reply on news.ycombinator.com