Cahn Ingold Prelog (Simon Proffitt) used AI to reinterpret the first 30 releases in the Waxing Crescent discography, purely using the album titles as prompts. At times, spectrograms were converted into audio, some tracks are raw, some are pure MIDI (based entirely on the AI output) and some are a mixture of both. There are no extraneous notes/rhythms, just what I was 'supplied with', albeit polished up a bit. In some cases the raw audio was good enough to stand alone. In some cases though the raw audio is boring, or annoying, so the next step involved converting it into MIDI info and then wiring the MIDI up to some virtual instruments.

Crescence process:

“Text-to-image diffusion models are trained to understand key visual concepts with billions of tagged images. It's a similar (but much more rapid) process to how we learn visual concepts. We know what cats are, because we've seen lots of different cats and understand their common features. And if we want to draw a cat, we use our generic knowledge to create a recognisable picture of a cat, but one that is (very likely to be) slightly different from any other cat picture that we've ever seen.

If the diffusion model is trained specifically on spectrograms, however, rather than pictures of cats, fruit, people etc, then the machine comes to learn that certain keywords have certain spectrographic similarities. It knows what a generic smooth jazz piece 'looks like' in the same way that it knows what a generic dog looks like. You can then prompt the machine to produce new spectrograms in any style you can think of. It has absolutely no concept of sound, only what sound 'looks like'. The interesting part, for me, is that it's unlikely that 'MondoProtozoa', for instance, is a concept that exists in the training data, so it uses its best guess as to what it is I'm asking for based on things that it does know. Some of these best guesses are pretty wild. It hasn't heard any of the music on any these releases, so any similarities are either coincidence or due to accurate titling on behalf of the original artist.

You then have to convert the spectrograms into audio, which I have a script for. In some cases the raw audio is good enough to stand alone - MondoProtozoa is pretty much just the raw output with a bit of reverb on. In some cases though the raw audio is boring, or annoying, so the next step involves converting it into MIDI info and then wiring the MIDI up to some virtual instruments. So some of these tracks are raw, some are pure MIDI (based entirely on the AI output) and some are a mixture of both. I haven't added anything else, so there are no extraneous notes / rhythms, just what I was 'supplied with', albeit polished up a bit.” - Simon Proffitt