Experimental platform using LLMs to generate algorithmic music | Hacker News Reader

Experimental platform using LLMs to generate algorithmic music

vibecompose.vercel.app

2 pointsmaddmann3 comments

I have been experimenting this past week with a small web platform that mixes JavaScript, music pattern generation, and large language models to create algorithmic compositions.

The idea is to have a language model suggest Strudel style music code which is then played back in the browser. It is still very rough. Mobile is not ideal, you need to enable audio or notifications, and the generated code sometimes breaks. Despite the bugs it can create interesting results when it works.

So far it seems to lean toward cinematic or piano style textures but occasionally produces unexpected rhythms.

I would love feedback from anyone interested in creative coding, generative audio, or live coding tools.

What directions would you explore if you were combining large language models and music patterns like this?

I just started looking at Strudel and was quite pleased to see this! I was impressed that it captured the idea of my prompt, which was fairly simple, but not too typical, I think (a brass call and response). What would be of most interest to me is to see how you are doing this, but otherwise, something to assist in learning Strudel. For example, take a more complicated piece from the Strudel website and help pick it apart and put it back together. In any case, thank you and I will continue playing with this while it's available!

Thanks for giving it a try! I am also a novice with Strudel so had a similar motivation when building this, and also just a curiosity if llms could output interesting compositions.

One thing you might find interesting is the search tool for searching strudels api and available sounds. That is available once you are in the code editor.

Lastly, one thing you could try is pasting in one of those complex songs and ask it to annotate it with comments.

Reply on news.ycombinator.com