You're right. I was confusing them. The disconnect between the discussion and approach to training music models and image models still applies, although the hypocrisy I implied does not. Harmonai are approaching it sensibly. Huggingface and most other image diffusion models developers are not.
You might be interested to know then that Harmonai and Stability AI (creators of stable diffusion) are tightly connected. I think Harmonai is actually part of stability, but it's a bit murky.