here is how music generation with AI works on my own machine :

1) On Jupyter notebook I execute previously mentioned couple lines of Python code, what starts web server that piers Web UI, runs both locally (in your PC) and in the cloud via proxy service (in Google Cola in my case)

2) in web UI I set the length: possible up to 300 seconds = 5 minutes = enough compared to 15 seconds in demos

3) I write the prompt

4) calculate estimation of generation time

5) I start generation

See: