We deliver a demo first. Clients can then request any changes before we lock it in as final. Clients can also send links to music they like and we try to achieve a similar sound.
It's a good point though - we work on a small scale now. We haven't thought through all the details once we start taking new clients. One free revision seems like our most likely approach.
If you have to employ composers, why do you need AI? I’m guessing you can hire 3 composers for the salary of a single ML engineer.
I'll hazard it's because they can produce more volume if they're just doing corrections and syncing