Suno (ai music generation)

[0bleak]

Has anyone been playing around with Suno (ai) to make music?
A friend has has a bit of success making some breakbeat hardcore/jungle so I thought I'd give it a try with making some weird hip hop.
The lyrics usually leave a lot to be desired.
It seems to take your instructions for the music to also use for the lyrics - I've even said not to make the lyrics self-referential then it comes up with lyrics saying they're not self-referential! Besides that, sometimes it comes up with lyrics but there are no vocals for the lyrics in the track, or the tracks gets cut off before the full track/lyrics finish.
Sometimes the descriptions for the tracks are kind of off - I'd rather it only use words that I've used for instruction.
Anyway, here are a few of the better ones (again, the lyrics and/or flows can leave a LOT to be desired)

Yo, check it, I'm about to drop some noi | Suno

experimental noisy avant-garde song. Listen and make your own with Suno.

app.suno.ai

(Noise, noise, noise) | Suno

experimental hip hop unconventional noisy song. Listen and make your own with Suno.

app.suno.ai

In this fractured world of noise and sou | Suno

experimental hip hop dissonant fragmented song. Listen and make your own with Suno.

app.suno.ai

 

[thirdform]

sounds like Frank Zappa lol. Not sure if that's a good or bad thing.

Will give it a spin later, see if I can get it to generate some decent hardcore acid.

 

[Mr. Naga Pickle]

sounds like Frank Zappa lol. Not sure if that's a good or bad thing.

Will give it a spin later, see if I can get it to generate some decent hardcore acid.

you've maybe already seen this, but low entropy's been writing quite a bit about AI assisted hardcore production lately:

How to create music using ChatGPT (Tutorial Video)

Remember our tutorial about creating music using Artificial Intelligence? - Especially regarding Hardcore Techno related genres. We turned ...

lowentropyproducer.blogspot.com

the tracks he's been making have been pretty ok!

 

[0bleak]

I found another one, cassetteai.
It takes longer to generate, but from the very small amount of time I've used it, it generally seems to be better.
Another advantage is that the focus is on the music so you don't have to worry about the shitty vocals and lyrics.
It's possible to pay for an upgrade to generate tracks longer than 1 or 2 minutes, and you can also do things like stem separation.
The downside is that it looks like you can only export if you upgrade, OTOH, it's just $3.99 a month.

 

[0bleak]

I just tried it and like suno, cassetteai doesn't much understand the concept of acid, but it's a little closer than suno although I guess that's not saying much.

 

[wektor]

musicgen is also ok. main problem with a lot of these big models is the diversity of the datasets. for example musicgen is trained on stock music so you can maybe do some good car commercial trance, but it does not even understand mainstreadm edm terms like hardstyle

 

[wektor]

you've maybe already seen this, but low entropy's been writing quite a bit about AI assisted hardcore production lately:

How to create music using ChatGPT (Tutorial Video)

Remember our tutorial about creating music using Artificial Intelligence? - Especially regarding Hardcore Techno related genres. We turned ...

lowentropyproducer.blogspot.com

the tracks he's been making have been pretty ok!

jesus christ he really needs to rethink his visual identification. and tbh symbolic generation for has been a thing for a while (like what, fucking twenty years or more probably?), if you wanna do "AI" and not "ML you even have the magenta VST toolkit which can write continuations of your midi, etc. and it's already pretty aged

 

[wektor]

 

[Mr. Naga Pickle]

jesus christ he really needs to rethink his visual identification. and tbh symbolic generation for has been a thing for a while (like what, fucking twenty years or more probably?), if you wanna do "AI" and not "ML you even have the magenta VST toolkit which can write continuations of your midi, etc. and it's already pretty aged

he's also written an ai assisted choose-your-own-adventure book abt being a gabber dj :

You are a Hardcore Techno DJ!

AI generated pick-your-own-destiny novel

youareahardcoredj.blogspot.com

 

[wektor]

he's also written an ai assisted choose-your-own-adventure book abt being a gabber dj :

You are a Hardcore Techno DJ!

AI generated pick-your-own-destiny novel

youareahardcoredj.blogspot.com

this is, excuse me for the lingo, horribly cringe

 

[Mr. Naga Pickle]

this is, excuse me for the lingo, horribly cringe

when u are being mean 2 low entropy this is who u are being mean 2:

 

[0bleak]

weirdly and coincidentally, I literally just got the Subcontinental Synthesis: Electronic Music At The National Institute Of Design, India 1969-1972 book in the mail - "The history of India's first electronic music studio founded in 1969 at the National Institute of Design in Ahmedabad by David Tudor."

 

[wektor]

when u are being mean 2 low entropy this is who u are being mean 2:

I consider some of his contributions pretty much invaluable but I don't think there's any reason to withhold from criticising this damaging kind of boomer aesthetics
like come on, we all have some awareness, and we all have internet
but maybe I'm so hard on it because I've been in the field for a few years now (even working professionally, since recently) and some of the ai aesthetics thing can be just overwhelmingly cringe, same as the crypto bro aesthetics.
but on the other hand, it will probably be the coolest shit around in 10 years and I will be laughing at this post, same as I do now at my memories of hatred towards frutiger aero

 

[wektor]

but yea bruh let someone younger or more visually aware do the label design for fucks sake, I can even recommend some people who would love to get on it

 

[Mr. Naga Pickle]

I consider some of his contributions pretty much invaluable but I don't think there's any reason to withhold from criticising this damaging kind of boomer aesthetics
like come on, we all have some awareness, and we all have internet
but maybe I'm so hard on it because I've been in the field for a few years now (even working professionally, since recently) and some of the ai aesthetics thing can be just overwhelmingly cringe, same as the crypto bro aesthetics.
but on the other hand, it will probably be the coolest shit around in 10 years and I will be laughing at this post, same as I do now at my memories of hatred towards frutiger aero

yeah, i'm just messing around. i wasn't a huge fan of this aesthetic back in the day and it hasn't really grown on
me since, but i do think it's cute that low entropy still finds it so inspiring. fwiw, i live in COOL MIDWESTERN US
CITY and a bunch of the 20 somethings i know reckon it's rad as piss...

 

[bun-u]

it's good for radio jingles - https://app.suno.ai/song/3c21269d-d496-4db4-8d6d-64ae1cfc0741

 

[wektor]

it's good for radio jingles - https://app.suno.ai/song/3c21269d-d496-4db4-8d6d-64ae1cfc0741

What I just realised listening to this is the instrumental on its own would sound like shit, or, even, like nearly nothing – who would've thought that the human voice is so much easier to model than music? And that we're able to accept really wide range of low quality audio, as long as there is a hi-fidelity voice on top of it

 

[bun-u]

I guess for established ‘genre music’ it just needs to provide some basic coordinates and your brain then fills in the blanks

 

[wektor]

this is actually a useful insight, need to keep it in mind

 

[echevarian]

I'd keep an eye on Hexorcismos and the RAVE model, I've been training one for about a week here.

That and a Musika model from scratch, which gave me stuff like this:


The artwork is just Midjourney, but I have finetuned Stable Diffusion models in the past and got much less "AI" looking results.