Using the web audio api, I can get an array of numbers which corresponds to the waveform of the sound an html5 audio element is producing. There's a good tutorial on how to do this. Then, using requestAnimationFrame (with a little frame limiting for performance reasons) I'm updating that array as the music changes. I then normalize the data a bit (or transform it slightly depending on the visualization) and redraw the screen based on the updated array. I'm using d3.js to draw and redraw SVG based on this normalized data. Each visualization uses the data a bit differently -- it was mostly trial and error to get some stuff I liked looking at. Since I'm using D3 -- which is just drawing SVG -- I was able to style everything in CSS (no images are used at all, including icons). There are a handful of differently colored themes for each visualization, and I do some rudimentary CSS namespacing by updating a class applied to the html element. eg. <html class='theme_1'>. This lets me override or substitute CSS rules pretty trivially. I can add some additional variation to each theme by messing with pseudo selectors. For example, I can use :nth-of-type to hide every nth SVG rectangle or making every odd child have a different stroke-dasharray, etc.