So here's how the song generator project is working out:
The generator is impossible. The method I decided to use (build a machine that understands and can orchestrate music theory) can't be done without massive amounts of music represented as machine logic.
Processing sheet music notation as machine logic is... actually not that hard, and it presents some really interesting possibilities for "smart music."
I'm exploring that right now, and it's really interesting.
Anyway, the goal with all this is to produce a new corpus of this stuff, that can be used against a conventional coding Lora, to help the machine actually understand music theory.
The reason why ChatMusician was never released is because the methodology doesn't actually work that well. Compositions are flat, boring, without a lot of thought.
Cool idea though.
And it would have worked if music notation didn't suck.
The problem is that music notation needs a major upgrade that positions it properly for the task.
So that's what I'm working on.
It's time to write another DSL! This is always fun.
Did a bunch of really boring foundational C++ work tonight. Made progress.
Sidenote: This would probably also work for vector images, but one thing at a time.