A little experiment in generating more interesting noise than standard perlin noise, by sampling the perlin noise function using itself. Most of the images below have higher res versions if you click on them.
No source code, since this was just a quick hack, and both code & ui are not presentable. If you want to replicate my results, simply use your favourite noise function (the colour images were made with turbulence on top of perlin's standard noise function, with fixes from this paper applied), then sample twice (in different areas or dimensions) using the original screen X and Y coordinates, resulting in new X and Y coordinates. Repeat this about 2 times more, and for each step use different scaling (you have to play around to get the best values), and in the last step sample R, G, B. Simple, but effective.
[UPDATE: here are a few more that use simplex noise instead, some with a fake “depth of field” noise function]