Music Videos are one of my favourite art forms, and that’s been true for a very long time. I faithfully recorded and watched and re-watched the Australian MTV, which was only on air once a week. I used to stay awake until ridiculous hours watching Rage on ABC. I don’t know art, but I know what I like. And apparently what I like is abstract, colourful, engaging, bite-sized 3 minute bursts of creativity.
Music videos are an art form that is surely unique. There are no constraints on music videos. They do not have to bear any relation to the music they promote. It is a field of complete creativity. It prizes innovation for its own sake. It also rewards doing something that has no particular formal artistic merit, it ‘s just “cool”. But it’s an art form designed for mass consumption. Art galleries are stuffed full of radical reinterpretations of what art can be. But it is exclusive. No-one sees it, and no-one needs to see it. Exposure is not really what it’s for. Music videos, by contrast, do get mass exposure, without having to conform to any kind of Hollywood standard.
These days I rarely notice new music videos apart form one source, Jamie Zawinsky’s mixtapes. He does have some kind of financial interest in promoting music generally, but it’s clear that he collates these mixtapes purely for the love of the medium. And I think that reflects what’s great about music videos: while technically they are a product of the capitalist system, nevertheless they exist only to be art, not to make a profit.
In 2018, for his 200th mixtape, he wrote:
Though I will eagerly argue that Russell Mulcahy both invented the music video and closed the book on the genre, if you were to make that claim, I’d be quick to tell you how wrong you were and point out that we are absolutely living in the golden age of the music video right now. The accessibility of the tools of production and distribution was supposed to transform the music industry into this cornucopia of new voices, and it so, so did not — the corporations locked that shit down tight. But for music videos… it kinda did… These days a band with $10k can put together a video that is every bit as high quality and insane as something that in the 80s or 90s would have taken $10M and a crew of 80, and that’s not just some tech-cheerleader cliché, small bands are actually doing that and it’s amazing.
I think that’s true.
And so, I think it’s a significant moment that a couple of months ago he included “The Blue Stone” in his 250th mixtape. It was completely generated by Google Veo 2, a deep learning model. It has a “director”, but ultimately, no-one actually made it.
In terms of content, it’s pretty mediocre. The core concept is soldiers dancing in ruins. This seems like a rather sophomoric idea. It seems like something that would appeal to people who liked to draw flaming skulls and swords when they were teenagers. And that’s me, I liked that. The video reminds me of another video posted a few months earlier, “Reckoning“, which I like precisely because it’s so over-the-top obvious, and that seems to be how it’s intended. I don’t think that’s true about “The Blue Stone”.
And it has the classic failures of these deep learning systems. First, everything is blurry, because the system can’t handle detail. Second, there’s no continuity. In particular, the skeletal jester figure mostly has two horns, but sometimes one, and sometimes three. Is that the same jester or multiple jesters? Arguably that’s not the point the video is trying to make. Further arguably, the video can’t be trying to make any point at all, because it’s a stochastic model. Arguably, why am I even watching this shit?
A couple of weeks ago it was time for mixtape 251. This includes “In My Heart“. And things get more interesting. I don’t know enough about these systems to identify which parts are purely deep learning models, and which parts are composited from that and other elements. But it has the same blurry, morphing artifacts that give away the involvement of machines. However, this time, the youtube description doesn’t give any credit to the system used to create it. It just has another man described as “director”. The video itself doesn’t seem, to me, to have any particular coherent message. But I’m a philistine, what would I know?
In the same set of videos there is “Lust in my Life“. Which seems to me to be a perfectly straightforward real music video, with a storyline and a message, and fitting to the music. But in the credits of this music video, alongside the usual artists, the director and editor is also credited with “AI generation”. And this is the point about “AI” in general. The idea of plugging your ideas into a machine and having it pump out what you want, that is never going to happen. The results are crap, because you need to work through your ideas and get to details. Without that, you haven’t created anything at all, and that will be clear from miles away. But as a way of enhancing and smoothing the work you’re doing anyway, it’s just a tool.
In fact the first deep-learning-generated video I saw was a collaboration between my old flatmates, from a year ago. At the time, I thought that was incredible, and I couldn’t understand how it was even possible (given what I know about the resources they had at their disposal). I quickly clocked that it was fairly standard machine-generated imagery. But given the musical genre involved, it was a perfect fit. The results are perfect, whether or not you can see the man behind the curtain. I can totally respect that.
“AI” is a tool, and tools will be used. I’m still bitter about autotune, but it’s clear that it is being used as a musical instrument to realise an artistic vision. It’s a shame that singers never need to learn how to sing. But that doesn’t make modern pop music worthless, and it certainly is not boring.
And it’s not like machine learning can ever come to be the only tool in the box. If I was going to give an example of the golden age of music videos, I’d offer “Our Ballads“. This is not even that technically difficult to accomplish. Technology made this possible on a shoestring, but it’s only with talented, committed artists that it can become art. Deep learning models have nothing to offer here.
However, I feel that the “golden age of music videos”, which really did exist, is probably now ending. What made that period special was the combination of many people being able to produce art, while at the same time very few people had the commitment to actually do so. This is what produced many works that were worth marvelling at. In a world where anyone can say “wouldn’t it be cool if there was a flaming sword, and it smashed into a skull, which was also flaming”, and that’s enough to make a music video… well, that’s a world with much more art in, but much less to marvel at.
And I do start to wonder: when the songs themselves are being generated by a deep learning model, will Jamie Zawinsky include those in his mixtapes too?