Reaction-diffusion models two virtual chemicals spreading across a surface and reacting where they meet: one feeds, one kills, both diffuse at different rates. From those few rules, pattern grows on its own, spots, stripes, tendrils, coral-like labyrinths, all shifting continuously and never repeating. Alan Turing proposed the mechanism in 1952 as an explanation for how animals get their markings, and the Gray-Scott model is the version every graphics programmer eventually implements.
The magic is that nobody draws the pattern. Tiny changes to the feed and kill rates move the output between dots, worms, mazes, and blooming fronts, and the parameter space is a landscape artists explore rather than design. On a GPU the simulation runs at screen resolution in real time, which made it a staple of generative visuals and demo culture.
It pairs naturally with music: audio can modulate the feed rate, inject seed disturbances on hits, or steer the diffusion, so the pattern becomes a living skin that responds to the track while keeping its own organic logic, close kin to the steered-physics approach of a fluid simulation.