Extension quality is the real differentiator for producers. Anyone can generate a catchy 30 seconds. Whether the ai music maker can continue that idea for two more minutes while keeping the same instruments and key is what decides if you can finish a song with it.
The chorus lift is my benchmark. Does the energy actually rise, with added layers and a wider mix, or does the ai music maker just get slightly louder? Real arrangement dynamics are hard for models and easy for a trained ear to spot.
.