Riffusion takes an audio file, converts it to a spectrogram (much like Capo can do) and then passes it through Stable Diffusion to make new spectrograms, and then back to an audio file.

It's like AI art, but for music.

I'm blown away by this, especially the sample where it makes a "beautiful 20-step interpolation from typing to jazz".

Here's some lo-fi beats for the holidays.

Sure, it's not amazing - but here's the thing; it's early days. And you know I'm absolutely going to use this to make music to play along with in the future.