Any thoughts on possibility of running the generation step on apple silicon via MPS instead of renting a cuda pod?

Sorry, I did not try it. Its worth a shot, I guess.

As a sound engineer, I can't believe what I see - all these sas tools are amazing. Maybe I missed it in the readme file: This requires a API key from generative AI models i suppose ?

Assuming you don't own a sufficient GPU, there are prescriptive instructions to rent a GPU from runpod.io

Ok thanks, now I understand "RunPod GPU" is a service. I simply read too fast.