Constraints that help
A blank canvas and an infinite parameter space produce nothing. Three fixed rules produce a series.
The rules that work are the ones that remove decisions rather than add features. One palette. One page size. No curves this month. Within a week the work starts arguing with the rule, and that argument is where the interesting pieces come from.
Parameters left open should be few and meaningful. Twenty sliders is not freedom, it is paralysis — and the pieces all end up in the same narrow region of the space anyway, because that is where the defaults sat.
When a series stops producing surprises I change exactly one rule and start again. Changing several at once makes it impossible to know what helped.