Would you score your visuals with music generated from rules instead of a model?
Music can be generated from plain theory rules, with no trained model at all, and it tends to sound structured in a very different way.
For your own videos, would rule-based music be a better fit than a track from a model, or neither?