In Transformer networks generating music, each colored line is a different attention head and the weight of the line is the amount of attention it gives to each location; each attention head looks for different kinds of patterns in the music, and the more attention heads a network has, the more powerful it becomes.