You can't mention "AI" and music together without a negative response. I think entrenched business interests as well as musicians and other music industry adjacent people, plus the well intentioned but poorly informed public have developed a visceral knee jerk reaction to the concept. Some of this is understandable, but I think it is mostly fueled by asrroturfing.

I am a musician as well as a technology enthusiast, and I think this a very exciting time!

To respond more directly to your point than your aside, there are a smattering of models out there that can take descriptions of sounds and do a decent job of creating them. (Stable Audio 3 just released last month and can do this, for example). - I don't find them to be useful for sampling, though. I'm still much quicker dialing in a sound with knobs or sliders than a text box.

Diffusion models in music making are not going away, though. This is (at least in part) the future.

For a taste, look at some of the interesting things being done over in the Demon project - https://github.com/daydreamlive/DEMON - to me, this is a much more positive use of the tech than "type words/get song".

Not quite what you suggested, but I did some experiments several months ago "enhancing" the samples in tracker music with some models, and they sounded terrible. There really is something about the sound of tracker files that's just right. But sure, you could generate lo-fi samples, there's a lot of computer generated samples in music, but putting them together into a pleasing combination is the hard bit.

An upvote from me, it's a reasonable question.

For your "search for bespoke and unique sounds", that sounds like a service Waves runs, called Illugen. It isn't built in to a tracker, but you can use it to generate the samples that you import into the tracker. Honestly I've never used it, I have too many samples anyway, but here it is: https://www.waves.com/illugen

But you also don't have to go down the generative / diffusion path. You can ask your AI to make a tool to generate various sample files for you mathematically. A good frontier model will happily make a program that will create some hihat and cymbal sounds out of white noise, a kick drum out of distorted sine waves. It will create some simple synth sounds from square waves. If you go deep enough, it will happily go down the rabbit hole with you into additive synthesis while nerding out about the Synclavier. Or it will do synthesis through frequency modulation while chatting about the Yamaha DX7 and don't do that because FM is like modular you'll never find a way out of that hole has anyone seen eno lately.

If you kept reading this far, there's plenty of Claudes already doing their own very simple synthesis & music writing, in videos like these below (two I found today). The models are capable of much much more if you give them time to build a toolset, and aren't just asking them to one shot an entire video about their existence using only a copy of ffmpeg:

https://x.com/nptacek/status/2065207326492020957/video/1

https://x.com/nptacek/status/2065230524264710313/video/1

Could you say more? I don't really follow, and I've used trackers for a long time. Don't some trackers already have something akin to this in terms of "randomizing" wave forms inside some reasonable parameters? Why would you need AI for this problem?

>Why don't we use AI to generate lofi samples for tracker music?

Because generating lofi samples is already pretty easy with waveform generators and existing tools. Burning millions of tokens worth of compute just to make a bass kick is profoundly wasteful.

look at inspect source. the comments like

    <!-- EXACT setup from working simple-test.html -->

    <!-- Chiptune3 now loaded via Vite build system -->

are pretty stark llmisms. also the random emojis in front of the duplicate and triplicate "Browse Music" links, the corny tagline in the footer, yeah nah. if this was made by human hands, it could only have been as elaborate satire. bad/inexperienced human code does not look like this.