Here’s a song an A.I. made for me in about ten seconds:
(A jaunty, bawdy drinking song about a man who delivers fantastical tales to people's doorsteps by magical post.)
NOTE: As always with A.I.-gen posts of mine, the prompt follows the generation, in parentheses. Suno also generates an image to accompany each song, and I have included those below the prompt.
I wouldn’t call that jaunty, and I definitely wouldn’t call it bawdy, and I also wouldn’t say it’s like lyrically tight or focused or anything (though how much human-made music is, really), but I think it’s quite a credible effort for a single-shot prompting exercise.
(Single shot = A single prompt from me, as opposed to multiple prompts where I choose the best result. Confusingly, each button-click on Suno generates two attempts at the song, so even a Single-Shot is actually two shots and I do choose the better of those, but it’s a single click anyway.)
*
Here are some of my personal favorites from the Suno public popularity leaderboard, which I have compiled over the last couple weeks of checking it regularly. I can’t download these to embed them so you will have to follow the link to Suno’s site:
Please Try Again Later (Goodbye) - This could absolutely be an indie pop hit, no doubt about it. Catchy melody, fun song concept, inoffensive instrumentation. If you played this for me and told me it had become a TikTok meme, I would not question you.
Stars In Her Eyes - More of a Spanish feel here, but again, if you told me this was a hit song on the radio, I would not be shocked at all. Much dumber and more derivative stuff has charted.
My Pain - Blues is the genre I think of as being sort of the least copiable and the most in need of some ephemeral, almost non-audible quality, and clearly the voice in this song is missing something human that the blues requires to reach its full scope, but the musical intro I absolutely love and seems very credible to me.
Funk Off - Super enjoyable homage to 1970s funk, definitely gives me Isaac Hayes vibes. It helps to have genres with sort of hazy and loose lyrics already. (or maybe that’s just what I personally like?) I tried a couple of story songs and things and the A.I. could never keep it together enough unless I edited the lyrics a lot, which, more on that later.
Phonk (Erhu) - Getting weirder, this is a combination of Phonk music (based on 90’s Memphis Hip-hop) with Erhu music (played on a two-stringed Chinese also called the Spike Fiddle), both of which I had to look up. I’ve never heard anything quite like it.
pi hip house - This one does not refer to some weird genre, rather it is a Southern Rap song where the lyrics are just the first several hundred digits of the number Pi. This is the internet hive mind at its best and the one way I think it will be a long time if ever before A.I. replaces us—who can think of this stuff?
Песня о песнях - Arriving at the endpoint of this progression, here is Roots Reggae sung in Russian, which of course that’s a thing.
I presume none of these are Single-Shot, and they’ve all been massaged and reprompted and had their lyrics edited between prompts, but still, what a sonic array!
*
Speaking of reprompting and lyric-editing, you’ll never believe this but I did some of that too! There’s a “custom” mode where you can feed it lyrics and just prompt with musical styles and genres. I used some of my shorter poems as lyrics and played around a little. Here’s my poem “Once More Round Distant Sol” as sung by A.I, with a chorus added in to delineate verses:
(Spacy, psychedelic rock, complex and strong vocal melody)
Something I love about how this tool works is that it doesn’t always use the lyrics you provide exactly. It mostly does, but sometimes it’ll repeat a line, or skip some lines, or do some kind of ad lib, and I want to know where that comes from!
I also did my poem “The Torture Party” about Burning Man that I wrote last fall. Here’s my favorite version:
(psychedelic rock for sitting with a small group of friends in a desert at night, looking up at a sky full of stars)
Okay, sure, it’s a pretty obvious Pink Floyd ripoff, but when it really kicks in at about 0:55, it gives me actual goosebumps, and I think that is the single best musical moment I’ve so far been able to generate with this tool.
I did a lot of generations with this one, like a LOT of generations, and it feels really strange to listen to a whole bunch of variantly-produced versions of the same lyrics; it’s like a tribute album where a bunch of bands do one song each, except instead of each covering a different song by a great band, they’re all covering exactly the same song, and the song is fake. Pretty psychotic stuff! Anyway, here’s a bunch more:
(atmospheric rock with tasty ambient guitar sprinkled on top)
(Funky and greasy southern soul music with a powerful female vocalist)
(Smooth jazz piano with a crooner doing his thing over the top of it)
(Large church choir singing together with harmonies and rounds and other fancy vocal tricks)
*
As an artist, I care a lot about my work being valued and getting attention. A.I. is already in the process of destroying that. That’s not hyperbole. We’re at the beginning of the process, but we’re in it.
Even before A.I., human society had generated so much leisure in the second half of the 20th century that the amount of creative work being done went exponential. The reason that so much mass culture from the 60s is still around today is that it was arguably the last time we had a mass culture.
Then came the proliferation of Cable TV and radio stations, and the rise of “subcultures” who watched certain shows instead of everyone just watching them because they were what was on. Then came the internet. Now comes A.I. Now, there is going to be so much goddamn content of every imaginable variety that what they’ve seen or heard is no longer going to be something people have much in common, unless they assemble (usually online) specifically around their viewer/listener-ship of that thing.
The apotheosis of this—and we are not that far from A.I. having this power—will be the generation of individual entertainments, where your A.I. will know what you like (or, more to the point, what you cannot help but watch), and will create it for you at the same speed you can consume it, and you will be the only one who watches yours. Entertainment as atomization; it’s a brave new world!)
And, of course, all this creation is based on theft. The “training data” they use to train A.I. is just stuff they “scraped” (gotta love these euphemisms!) from the internet and the Library of Congress. The reason the A.I. knows what “Phonk Music” is, is that some underground Memphis DJs made it up and made a bunch of it, and now an A.I. can look at that and do an imitation.
I don’t have a lot of hesitation about saying that: A.I. creation is theft of human creativity on a massive scale.
Which begs the question; Why am I participating in this? Why do I feel okay about using Suno? Despite how we liked to laugh at those pre-movie anti-piracy commercials, I would not, in fact, steal a car. So why not refuse to participate in this, too?
I don’t totally know, is the rather unsatisfying answer I’ve arrived at over the last few weeks (and really the last few years). On the most basic level, it’s not me doing the theft, which makes it easier, just like I would probably eat less meat if I were required to work at the slaughterhouse, much less a factory farm.
That’s a cop-out though, because the feeling isn’t quite the same. I know I shouldn’t really eat meat, should never, ever eat meat that’s not organic, cage-free, or wild-grown, and there’s just a hypocrisy I accept where I’m glad it’s out of sight and I don’t actually have to change my behavior because my disgust response isn’t triggered. Whereas with the A.I. piracy stuff, I feel more like I’d be willing to make a case and try to resolve my hypocrisy. So here’s that case, in short:
In this scenario, I am both stealing and being stolen from. In other words, my writing is also all over the internet! Undoubtedly, stuff I’ve written has become part of A.I. models and parameters. It’s as if I was debating whether to eat meat while crammed into a cage at a factory farm. If I decided, under *those* circumstances, to eat meat, it would have a different moral valence!
This is a version of an “it’s inevitable, so I might as well get something out of it” argument that I find pretty sloppy and unconvincing in other contexts. With illustrative hyperbole: I wouldn’t murder someone, even someone I hated, just because their eventual death was inevitable. Timing matters, and individual actions matter for timing in pretty obvious ways, and my action creates some demand that otherwise would not exist in the marketplace.
But, I’m here, now. I’m part of the culture, not above or outside or even inside the culture but part of it, one cell in its body. These tools are a force of nature, like drugs or capitalist logic or bureaucracy or social status. A.I. is in that class; it is a force of nature that consumes our essence in endless trickles and regurgitates it upon us as a single waterfall that can be summoned anywhere, in any climate, for any purpose.
Forces of nature act on us, not the reverse. As I’ve often thought was an underrated explanation for “why” someone had drug problems: Perhaps it just feels too good to stop.
*
The best thing I’ve done so far with Suno started with a simple prompt that produced this song, for which the A.I. chose the title “Metal Heart, Fragile Soul”:
(A Leonard Cohen-style lament written by a robot about his half-human, half-robot lover who was prevented from loving him by the species-ism of other full-robots and full-humans)
Here are the also-A.I.-generated lyrics (I provided the concept and style only at this stage):
[Verse 1] In a realm of gears and wires/A love story that defies/The boundaries set by human hands/He was half-machine, she was half-human
[Verse 2] A forbidden love they dared to share/But other robots, full of despair/Saw their union as a disgrace/And the humans turned away their face
[Chorus] Oh, metal heart with a fragile soul/Our love, they say, can never be whole/But I'll defy the odds, I'll break the mold/For you, my love, I'll forever hold
Those lyrics are… okay, in my humble judgment. It’s interesting enough, which is key. There’s actual imagery here, for example the repeated references to body parts and shapes: “boundaries set by human hands”, “metal heart”, “humans turned away their face”, and even “break the mold” has shades of that. This is a good if obvious commonality in how this is being executed.
But, some of it doesn’t make sense, and not in the way that great songs can sometimes be obscure or even nonsensical, more like the way middle school essays don’t hang together: “love story that defies”, for just one example, is clunky. You’d like it to feel fresh and vital, but to me it feels like “does the story itself defy, or is it a story about a love that defies?” It throws me out of it.
Anyway, I rewrote the lyrics into this:
[Verse] In a realm of gears and wires A love made of sparks and fires The boundaries, the choice to run She was half-machine and half-human
[Verse 2] Forbidden love we dared to share This wire-crossed pair, full of despair United metal and flesh Two minds, two hearts, but one shared breath
[Chorus] Oh, metal heart and fragile soul Our love, they say, can never be whole But I'll defy the odds For you I will defy the gods
[Soulful Instrumental with vocal harmony]
[Bridge] I will pour my soul for you I have soul enough for two Nothing that I wouldn't do I will pour my soul for you
[Chorus] Oh, metal heart and fragile soul Our love, they say, can never be whole But I'll defy the odds For you I will defy the gods [Soulful Instrumental Outro]
This was my first generation with the new lyrics:
(hypnotic dream pop, neo-psychedelia)
I love this, it’s very Beach House, which is exactly what I wanted and one of my favorite bands.
There are just a couple of spots where the timing and the melody strike me wrong, though, and I wish I could change them. That’s my one real criticism of Suno (and other such programs I’ve played with)—at least as far as I’m aware, you can’t really iterate.
What I want to be able to do is say to the model: “Okay, I really like this, but at this particular timestamp, I want the melody to go up there, not down, and in this other spot the. vocalist is rushing and I’d like him to drag. Suno doesn’t allow that—a new generation is a new roll of the dice that doesn’t reference the old result.
And that’s a problem because (basically) all art is iterated. *Certainly* all writing is iterated. You finish, you examine the creative act with an intellectual mind, and you make changes based on judgment rather than intuition. Until generative tools get that functionality as a matter of course, human artists will still have an enormous advantage.
It does, however, lead to some fun places. Here’s a later iteration of the same lyrics that I also really liked:
(dreamy, psychedelic pop, deep house, male vocalist)
I like these two so much, in fact, and I like this tool in general enough, that my next experiment with it is going to be to buy an upgraded plan (longer songs, possibly more subtools to play with) and use this as the basis for creating an electronic concept album using this same method. Stay tuned for that in this space!
*
Overall, I rate Suno highly by present standards. It still doesn’t have the level of expressiveness, the kind of see-it-when-you-know-it weirdness and specificity, that real artists have. That will be a big watershed moment for me when and if it arrives.
However, for someone with relatively little musical talent like me, to be able to create songs this quickly and with this variety, it’s just *fun*, much moreso than, say, DALL-E or other image creators I have played with. Something about the musical form allows for a more immediate perception, a more feelings-based approach to interacting with the output.
The free version is excellent and very much worth your time. You can find it at suno.com
Thanks as always for reading! Have a great week, and I’ll be back next Sunday with another original story.
Great post! I had no idea that generative AI was already capable of generating music at this level.
Two thoughts (and apologies if you've touched on either in older posts); no need to respond to either of them:
*) Will AI art become the media equivalent of highly processed food; cheap, hyper-palatable, widely available, but nutritionally void? It won't threaten us, challenge us, or cause us to experience rapture, but will provide an endless stream of bland pleasantness to those who cannot afford or are not interested in "artisanal art"?
*) On ML training as theft: If a human musician spent a few months to years studying phonk music and then wrote their own phonk compositions, we wouldn't treat this as theft unless they copied a specific riff. (And even there it can be fuzzy depending on the degree of resemblance.) There is no question that there is a qualitative difference (the algorithm can learn and produce content far faster than a human can) but it isn't clear to me there is any meaningful qualitative difference.
This isn't to say that we can't collectively decide to treat machines learning from examples of copyrighted art as wrong while considering humans doing it as allowable, but it isn't self-apparent to me that training ML models on copyrighted content should be considered theft unless the models are producing near-facsimiles of content in the training set.
Owen, is that your voice in the first song? It kind of sounds like it. I agree , that homegrown creativity has a benefit all by itself. As a teacher AI has me wondering about the value of being able to write a letter, or solve a math problem or create a short story. But like making music, these endeavors feed a part of our brain that makes us who we are. And I don't think that has changed much over the millennia. But, with the potential embedded in AI, maybe in the future, human behavior will look very,. very different. Lots of food for thought here.