Some things that set this one off -
-There's a basic particle generator that's being "mandala'd" via a core image kernel.
-The motion of that is then run through an optical flow image process, that calculates vectors of movement in the image. Based on that, gl points are placed (the tiny particles).
-There's some feedback stuff involved in the look of the particle.
-For something a little more refined looking, the entire generated visual is run through a realtime process that approximates something like a lens blur, using a core image kernel to run a fragment shader.
-Audio peaks influence particle amount, while spectrum influences color.