tfw when you have to turn on your VPN to view an imgur link because you're a UK pleb.

Thanks for the wider context of the discussion.

I see what you're saying about Suno's usefulness as a tool.

If the whole point of it is to be able to output sounds that are tailored by the user's input, it is a failure. Or at the very least, my prompt goes a long way past its ability to do so. It's probably not even trying to follow my prompt text and instead just reacting to the genre tags.

Not that I wasn't aware of this before you pointed it out. I was disappointed that it didn't meet the prompt, yet was still intrigued by the output. And I essentially discounted this in my head by thinking: "well, often my own music ends up far away from original intention too."

But yeah, ultimately it could be argued that the most my prompt proves is that Suno does not have the ability to synthetise more disparate, potentially interesting/unique combinations of musical ideas. And it does not have the ability for people know who actually know about music to meaningfully influence its output in ways they might wish.

Instead, it just says "nah" and spits out whatever it wants from its training data.

And seeing as though Suno is supposed to be some all-powerful generative music AI powerhouse, if all it's able to do is regurgitate known quantities then how much artistic value does it really have? Arguably very little.

What's more, the majority of people who are using it aren't even at least trying to achieve something interesting. Instead, they're genuinely thrilled with the regurgitation they're fed (and which they asked for).

I mean, this first example on the Suno homepage as "Best of V5": https://suno.com/s/4dzYSNAQ9CVFrck8

I can accept that. And it does change my opinion towards it.

I see too what you're saying about it being an endpoint.

I was going to make the argument that even if it's an endpoint, it's still a potentially useful tool for creating those endpoints and using them as a jumping off points for whatever it is you want to create. But the reality is that's what humans have been doing for the whole history of music anyway, just with other human's music. So in that respect it continues to provide no new value whatsoever.

Ugh, fine, I guess I'll go back to making own music then. I hope you're happy.

At least I've got a new tool to check out, Scapeshift looks pretty cool. Thanks for sharing.

I'm far from as storied and experienced as yourself, but I've got some credentials.

I have a masters in music composition. I too spend hours every week scouring for new music. I've made music that's convinced someone to spend thousands of their hard-earned money to commit it to vinyl. Over my compositional lifetime, my music has had 1 million + streams

I know that's small fry in the grand scheme of music, but still more than the majority of other musicians on Earth can say.

Is this Hubski Ghosts song shite? Yes.

Outside of any comment of the model's ability to produce something that sounds like music, the generation itself is slop. I agree.

It's output is technically competent but has zero artistic value. I'm not about to call you a prick for pointing that out.

However, I also don't feel that it's epitome of what Suno can produce. Far from it.

I've spent a fair amount of time playing around with Suno since it first came out. From making it glitch the fuck out to stretching the limits of the model's ability.

And now? I genuinely believe it can actually generate interesting things.

Take this example: https://suno.com/s/KvNR6LGu4ftcIZrO

Now, you may like this style of music or think it's just some jazzfuck shit. But do you honestly not see anything impressive in it?

Yes, the fidelity is questionable. There are clear artifacts in the audio. But to my ears there's a variety of interesting musical features in it.

It's sits within a genre I listen to a lot. I have gone back to listen to it again. I can recall things about it. I find it hard to say it's not in any way good.

Now, is any of it actually more impressive than what the best musicians have created No. I'm certainly not here to claim that this generation means Snarky Puppy should give up or that the top 40 is soon to be all AI.

I don't know, perhaps my sense of judgement and quality is fucked or I don't get the distinction between "things I like" and "things that are good".

But I can't believe anyone can listen to this and not have even one small positive thing to say about the model's output besides that it can adequately generate music to the level of bland, forgettable stock music.

What am I not hearing?