Suno V3: Song-on-demand AI is getting insanely good

It's the end of the world as we know it, and I feel like making some AI-generated tunes. Massachusetts-based company Suno has released its third text-to-music tool, and it chops out fully-produced two-minute songs in any style, complete with vocals.

As in every other human endeavor, so in music. Our own Rich Haridy's words from 2017 are starting to look remarkably prescient. "At the rate AI-generated music is progressing," he mused, "it could be old-fashioned to manually compose songs in as little as a decade."

He was writing about the now-defunct Amper music composition tool, a professional-grade and reasonably complex system that would generate high-quality single instrument tracks for music producers to take away, stick in a digital audio workstation (DAW) and edit up into a song, or a backing track that a singer could then perform over. The results were incredible.

By this time last year, Google's MusicLM signaled that mixing and production were next on the list, making short clips of instrumental music in which the entire band, orchestra or ensemble were all generated together. And a whole range of other AI startups were starting to have fun with vocals – letting people digitally transform their voices into other, more famous voices, for example.

Suno is one of a different group of AIs that simply do it all for you, in a single shot. Type in what kind of song you'd like to hear, and it'll generate a couple of options for you in about 30 seconds. You can give it the vaguest of directions, or be quite specific. You can type in your own lyrics, or just give it some idea of how you'd like them to feel. You can talk about what kind of instruments you'd like to hear, or let it surprise you with the style it comes up with.

So let's hear how it goes, eh?

This one's interesting, the lyrics are written by Anthropic's Claude 3, taking a swing at OpenAI's Sam Altman and the GPT model.

There are still occasionally some weird artefacts in the audio, it has a definite preference for auto-tuned voices, and you can rest assured the lyrics it comes up with will be GPT-level cheesy unless you give it some solid direction. You'll be able to pick these AI-generated songs in a number of ways, especially by their tendency to shoehorn lyrics into a form in strange ways.

Still, when it nails a song, it's already getting hard to tell. It's remarkable to see how many aspects of music this thing has swallowed, digested and is now able to regurgitate. As a simple consumer app, it'll generate two versions of your first five song ideas for free, then give you 500 more for US$8 a month. You couldn't even get a jazz musician for that.

As with all these AI projects, Suno V3 might be surprisingly good, but it's also the worst it's ever going to be. As a former, and still aspiring musician, I'm not sure how to feel about it. In some sense it makes my skin crawl to think of the corners that musicians will soon be painted into.

On the other hand, high-quality personalized songs on demand, about any silly topic that pops into your head... That's a lot of fun to play with. The #1 song on the Suno public charts at time of writing is a punk-metal opera whose lyrics are a recipe for bean soup. It's fantastic. And in terms of idea generation, it's pretty awesome to hear things fully realized like this, production and all.

I don't mind the idea that if it's been done before, you might just as well let AI do it – maybe that leads to new realms of human creativity and the elevation of music that's more adventurous... At least, until the AI instantly subsumes those new ideas too. And it's hard to see the younger generations working hard enough to master an instrument, when they can get the vibes they want at the touch of a button.

Perhaps human music leans in toward collaboration and community, where it came from? It's hard to imagine AI being able to move air and move bodies like a great live band, but then it was hard to imagine AI getting to this point so quickly as well. Either way, there's no stopping this wave, so we might as well find ways to enjoy it.

Source: Suno