The first time I tried an AI music generator, it was… amusing. And just that. It was interesting to watch a model attempt to make music, and the output was impressive once you reminded yourself a robot made this. But generating music was only ever going to be for my own pleasure — it had no practical value — so I forgot the technology even existed. It also wasn’t remarkable enough to stay in my mind.
A lot of time has passed since then, and not long ago a friend played me some AI-generated music. That was enough to jog my memory and send me back to check these tools out again. Dare I say, AI music generators have improved way more than AI image generators or AI chatbots have. Let me show you.
AI music got scary good while we weren’t looking
“wow, a robot did this”
The first AI tool I tried was Suno. I messed with it in early 2024 (around January), and later I wrote about how to use it to make your own song. In between, I tried Boomy — another AI music tool. The difference between the two was pretty clear: Boomy often gave you a “better” end result, but it gave you very little freedom over what the song should sound like. Suno gave you a lot more freedom (you can type in whatever you want), but back then it was much more likely to spit out incoherent slop.
There are other tools too — like Google’s AI instrument playground that let you sample an instrument and play it right there. You can’t generate full songs with those, but you can use them to make songs.
Anyway, back to Suno. Before I lose your interest, let me show you what I mean when I say they’re absurdly good. Take a listen to the song below:
I made this in about 10 seconds, without installing anything, for free. This song is entirely AI-generated. If it sounds familiar, it’s because I fed it an edited version of Colter Wall’s Nothin’ (which itself is a cover). It’s good, isn’t it? And it’s good in a slightly terrifying way, because I’m not confident I would’ve been able to tell it was AI if you didn’t tell me. That’s where we are.
You can feed it just about anything. All of this is near-instant generation. I didn’t even bother writing a detailed prompt or asking it what to do. I just paste lyrics and add a few style keywords. But it can do a lot more when you take the time to guide it.
Can it make indie rock?
With a couple of keywords…
OK — let’s try indie rock. “Indie rock” isn’t exactly a precise genre, but… that’s true for most genres, isn’t it? Either way, I can usually steer the vibe by throwing the right keywords into the Styles box. I’m not a musician, but I’ve listened to a lot of music. I can usually tell what ingredients I’m hearing, and I can toss those ingredients into a prompt.
Styles aside, another important parameter is the guidance you provide in the lyrics box. I remembered from my first time using Suno that you could add directions in brackets. It won’t sing those, but it’ll treat them as stage directions — at least, it tries to. I say “tries” because it doesn’t always stick to them. Take a listen to the example below:
In this case, I already had the seed of a song. I’d written bits and pieces of it before, so I threw it at ChatGPT and asked it to turn it into a complete set of lyrics. Then I added bracket guides and fed that to Suno. In the style box, I used these keywords:
indie rock, alt-rock, post-punk-leaning drive, heartland/anthemic indie, slow-burn to big chorus, late-night confessional tone, clean electric arpeggios, palm-muted eighth-notes, chorus-y shimmer, edge-of-breakup overdrive, amp tremolo outro sustain, hook motif tag, brushes-to-rim-clicks build, half-time final chorus, steady kick-snare anchor, walking bass lift, tom accents, close-mic spoken vocal, small-room ambience, distant synth pad, dynamic quiet-to-wide arrangement, subtle slapback/short delay
If you listen closely, you can hear a lot of that come through. I’m pretty happy with how it turned out. In fact — dare I say — it’s kind of a touching song.
And that’s where the weird questions start. Art is anything that evokes emotion. The emotional core here is the lyrics… which were also generated by AI. Sure, I gave it direction, but the final result was still “written” by a model and “performed” by a model. So what does that mean for art? Is AI able to evoke emotion?
Can AI make techno?
Dark, gritty, industrial techno
Techno is a tougher target. There are no lyrics, so evoking emotion is entirely on the instruments. Without words, it needs to be really good to hit you in the chest and qualify as anything more than “a loop that goes hard.” I have a lot more respect for good techno than good rock for that reason.
So this is a tougher ask from AI, and it’s also more dependent on my guidance. A sinister melody, an anxious cutoff rhythm — those ideas go a long way, but when I type those words in, it’s still up to the AI’s interpretation of “sinister” and “anxious” to actually deliver on them. Here was the result:
Aside from a paragraph of style keywords, here’s what I put in the lyrics box for this one (since it had no lyrics):
[Tempo 145 BPM, 4/4, gritty/industrial palette, tight arrangement, moderate sidechain on rumble]
[Intro: warehouse room tone + filtered noise, sparse metallic ticks, HPF slowly opening]
[Build: ghost kick (sub muted), rumble tail fades in (LPF), short riser from field-recorded whine]
[Drop 1: full distorted kick (hard clip) + mono rumble, closed hats 1/16, metallic hit every 2 bars, atonal stab motif (repeat)]
[Variation: add FM growl layer on fills, 1-beat mute every 8 bars, hats slightly wider]
[Break/Reset: kill kick, keep drone + granular noise, gated reverb snare pulses, quick filter sweep down]
[Build 2: kick returns filtered + quieter, short snare roll (metal hits), last beat: stop]
[Drop 2 (peak): kick + rumble harder, offbeat ride/open hat, second stab layer higher/shorter, noise bursts on downbeats (sparingly)]
[Outro: strip to kick + drone, fast LPF to room tone, end on short reverb tail]
I’d say the results are good. I wouldn’t call it dark techno, though. It’s generic — but it’s good generic. If it came up on shuffle, I probably wouldn’t immediately clock it as AI. But I also wouldn’t pick up my phone to look up the artist.
Compare this to something like A Burning Question by Tim Tama and you’ll see the gap. Even if they share a similar palette, Tim Tama manages to aim the emotion. The AI track feels more like untargeted anxiety: it moves and churns, but it doesn’t really say anything.
A banger I won’t listen to twice
Yes it can. Frankly, this is where my new (and probably short-lived) obsession with AI music began. A friend of mine was insisting I help him get lab mice — just one pair — to keep as pets. He’s really into animals. They don’t sell lab animals to civilians (for ethical and environmental reasons), but my friend kept insisting anyway, so I replied with… a song.
Before you listen, here’s context: most of the metal I’ve listened to has been Dethklok. They’re a parody band from a parody show. I genuinely love their music, but I’m aware this isn’t exactly “serious metal.” So if you’re a metalhead reading this, I beg your pardon.
It’s quite the song, no? I love it. It did what it was supposed to do — although I can’t listen to it twice without getting a headache.
My final take on generative AI
Mediocre isn’t useless
I’m a writer, so I know the “AI will take our jobs” fear as well as anyone. But I’ve grown to appreciate AI. Not because it won’t replace jobs. Oh, it will replace some.
AI is getting very good at giving you the mediocre version of almost anything. And it’s mediocre in a very specific way — it tends to regress toward the most average, broadly acceptable output of what it’s seen. Mediocre isn’t automatically bad. Sometimes “good enough” is exactly what we need.
AI won’t replace writers in the way people dramatically imagine. What it will replace, though, is a lot of copywriting. AI won’t replace artists as a whole, but it will replace the commission painter who paints your portrait for a fee and does just that.
To put it in perspective: AI is replacing Fiverr. You know how there are literal bands and musicians on Fiverr where you can pay them a hundred bucks and they’ll sing anything and turn it into a song? That’s the kind of work AI is eating first. There are copywriters you can commission to rewrite your cover letter for a university admission. That’s the kind of work AI has already started swallowing.
That’s my two cents. I know AI music, AI art, and generative AI in general are touchy topics — but I hope you get what I mean here.
