For once it's nothing like LLMs and image diffusion models we already have. It doesn't let you generate for any keywords that represent names, bands, brands, etc. So there no let's say easy way to compare what it generated with something that already exist. Not even Beethoven.
And while it can generate something when you ask for national anthem result is quite underwhelming. It can't generate music for specific musical instrument either.
What it good at is at somewhat following instructions: pacing, trying to fit specific mood. I managed to make it generate some Synth-pop as this subgenre less affected by ambient noise.
