Bah, the user interface and design is horrible to navigate in, which just made me sad. Try to paginate, music stops etc. If you created this, spend some more effort in user testing before sharing if you want users to have a good experience. My 2 cents.
Ngl the weird UI bugs made me think it must be made by AI, either that or the developer has very skewed frontend skills where gradients are fancy but sliders & interactions are broken.
I am actually going to guess that this is in fact unfortunately not an LLM. I have been made a few audio webpages and don’t think a lot of these mistakes would have been made, even though the model can’t hear :^)
<!-- 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.
> <!-- EXACT setup from working simple-test.html -->
All LLM-kind would be vastly improved if the words "exact" and "brilliant" were nerfed to hell in their pre-training weights or even just removed from their training distributions entirely. Virtually nothing outside of mathematics is "exact", and virtually nothing outside of colors should be described as "brilliant".
Vibecoded website with poor UX. Loving that the website is both trying to be fancy by having a floating player you can drag around with a playlist, while also wiping everything if you click the wrong link. No human made this, or paid it any attention at least.
> No human made this, or paid it any attention at least.
I think we’ve got to get used to seeing those as the same thing. Paying attention to it is making it, in essence, the more attention you pay, the more you own the process, the more the result is ‘yours.’
I wonder if you can find a way to turn the device volume up to max to simulate the unexpected music blasting out and surprising the hell out of the user.
This is a nice collection from the ReclusiveLemming YouTube channel [0] . The video description contains download links to the original tracker modules and an MP3 mix.
I used to listen to these modules with Xmp Mod Player [1] from F-Droid on my commute to work with my Nexus 4.
Many of the modules contain "hidden" messages that the Android app made easy to read.
I tend to think this song is the magnum opus of keygen music. Always used to be associated with Sony Vegas tools which, when I was younger, was a big deal cracking.
Horribly coded site but a cool collection of music, only wish I had access to the original collection because I know without a doubt that they just downloaded those mp3s from someone else’s site.
>Preserving the digital underground's musical legacy
Yeah I'm pretty sure the people who made keygens and chiptunes would hate today's AI and LLM. This is a tribute to chiptunes and keygen music as much as putting a picture of your grandma into an AI tool to animate it would be a tribute to her.
the demoscene is about putting in lots of technical effort into programs even if it would be completely unreasonable in any real software project.
This vibe coded mess is putting in so little technical effort even though it is completely unreasonable for any piece of software associated with the demoscene.
I think maybe this is getting hugged to death. I searched for an old favorite of mine: `radix - bright eyes`, and couldn't find it, but maybe I'm just doing it wrong.
Not that I've looked into it much, but a thought just occurred to me. Why don't we use AI to generate lofi samples for tracker music? Why aren't there trackers with that feature bolted on? I should be able to search for bespoke and unique sounds out of thin air.
Surely that should be a very modest goal to achieve?
(re: downvotes... I say "AI" as a synthesis method, not as a way to interfere with the creative process, but I guess I have to resign myself to the fact many downvoters might be ignorant of how these musical sausages are usually made)
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.
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".
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:
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.