Urban75 Home About Offline BrixtonBuzz Contact

Text based Music AI programs - AI making music

ska invita

back on the other side
Its just dawned on me just how much we are living at the start of a new technological revolution - AI is about to disrupt everything.
Heres a little video about Googles MusicLM.


The video reckons this MusicLM is the best known so far. IT is very impressive. A little way to go but the speed of improvement in AI and the billions being pumped into it all it wont be far off before its hard to tell the difference in some genres.

There are other type in text>Out comes music programs out there - anyone comes across one we can access?
 
Heres one you can play with - not as good but does work - sort of, not really

that google one is a different level

I suspect there’s just a random number stream and a bank of samples underneath all that, and the input word is just a seed.
 
its already hit next level - have a flick through this for examples...and then compare to examples at start of thread not even a year ago

 
Last edited:
I think it's excellent that anyone can make music now, without spending fortunes on kit.
tbf "youre" not making music, the AI is... its just inputing text for the human....a bit like commissioning music from someone...that someone being a robot
 
tbf "youre" not making music, the AI is... its just inputing text for the human....a bit like commissioning music from someone...that someone being a robot
That's similar to what was said about synths when they came out. Some people still insist synth players aren't 'real' musicians.
I don't really see much difference between someone who can't sing, getting autotuned to acceptability, and someone who can't sing, typing instead. If you make music, you're a musician, and that AI music does sound quite realistic. All they have to do now is get the drums completely out of time and they've nailed it. :D
 
That's similar to what was said about synths when they came out. Some people still insist synth players aren't 'real' musicians.
I don't really see much difference between someone who can't sing, getting autotuned to acceptability, and someone who can't sing, typing instead. If you make music, you're a musician, and that AI music does sound quite realistic. All they have to do now is get the drums completely out of time and they've nailed it. :D
good trolling :thumbs:
 
Spoke to someone who said it'd been their birthday this week and their manager played a three-minute AI-generated rap in a team meeting to mark the occasion. If you want a vision of the future, imagine your manager playing an AI-generated rap about your birthday in a team meeting forever.
 
Spoke to someone who said it'd been their birthday this week and their manager played a three-minute AI-generated rap in a team meeting to mark the occasion. If you want a vision of the future, imagine your manager playing an AI-generated rap about your birthday in a team meeting forever.
Welcome to the future.
 
If all you want from music is something nice coming out of a speaker, then AI will do the job fairly soon, if it isn’t already.

But if that’s all you think music is then I genuinely feel sorry for you.
 
If all you want from music is something nice coming out of a speaker, then AI will do the job fairly soon, if it isn’t already.

But if that’s all you think music is then I genuinely feel sorry for you.
if it puts lots of industrially produced pop pap out of business that would be a wonderful development
too much to hope for though i expect
 
I think the next step is creating tunes from ideas (like the humming example mixed with text) and then actually being able to create stems which can be directly plugged into a DAW for human input and manipulation.

Having these things spew out an mp3 of a finished article is only so useful. If I’m a musician/producer wanting to create with AI - I want it to be a tool, not the be all. I want to be able to break out the constituent parts and cherry pick the elements I like, toss the bits I don’t like, and have some human input with the final mix.

*this may already be possible but I haven’t seen it yet.
 
If it's anything like AI art...where people end up with 6 fingers and multiple legs or arms..then I wont be listening.
 
I think the next step is creating tunes from ideas (like the humming example mixed with text) and then actually being able to create stems which can be directly plugged into a DAW for human input and manipulation.

Having these things spew out an mp3 of a finished article is only so useful. If I’m a musician/producer wanting to create with AI - I want it to be a tool, not the be all. I want to be able to break out the constituent parts and cherry pick the elements I like, toss the bits I don’t like, and have some human input with the final mix.

*this may already be possible but I haven’t seen it yet.
Google made a version of this a few years ago



I used it recently. As well as the use case you mentioned, with humming -> instrument, you can also use it like I did to 'humanize' midi instruments.

I made a saxophone riff and then sent it into the plugin set on saxophone, too. The output was a more natural sounding sax, rather than my crappy rigid midi version.
 
We've got a show coming up, it's a deep-listening project. The soundtrack design has been written by sound-designing humans primarily, but as the soundscape is on a 20minute loop, but each performance is slightly longer, the infill moments will be autogenerated utilising Bronze AI, an animative music composition tool.
I'm intrigued by this and may sit in multiple shows and try and work out what I can hear is different
 
I'm impressed by Udio. I asked it to make a Kraftwerk tune, and it sounds like Kraftwerk. I asked for a Jeff Mills tune. It didn't sound like Jeff Mills but it's MUCH closer to making plausible techno than anything I've heard from these AI music things so far - the others have been absolutely shite.

Now I've just asked it to make a tune like Moodymann. It came up with this:


Definitely not Moodymann quality - way too generic sounding, but it did this in seconds, and without me actually mentioning Moodymann directly (there are flimsy copyright protections, meaning you have to describe him rather than using his name).
 
They're oversubscribed, but someone has already done the only thing I could think of trying - albeit I don't know what style I would have chosen..... and using a different platform.



Far from the first time too !


I think they've made a major fuck up with the launch. I have 600 generations per day for free.

I'm sure this won't last and they're using everyone who has signed up as testers, but that's way too much, IMO.
 
Back
Top Bottom