in reply to Randomly biased, random numbers.

After skimming through the thread, my idea is:

  1. Generate a density gradient image with the clouds render in gimp/photoshop.
  2. Apply distortions for flavour.
  3. Save above as a macro so you can generate lots of distributions with one button press.
  4. Save as greyscale BMP
  5. For each pixel, convert it to a sample point if brightness > rand(1000)

Note: I presume you don't need an exact number of samples, but if you do, just shuffle the list and then pick the first N.

EG:
http://i.imgur.com/OvPWjWf.png - GIMPed Difference Clouds + more contrast + gamma correction for darker darks
http://i.imgur.com/Di5wzG4.png - Resulting random samples. (about 3000)