I'd like to implement a system that generates unique NxN blocks when given a MD5 hash as an input, currently I'm splitting the MD5 into sub-strings and just using them as Hex colour-codes, does anyone have any good ideas on systems I could use to generate these images?
I've considered using the values as the constants in a Lorenz attractor but the output just isn't that visually distinguishable for large numbers of generated outputs.
The auto generated avatars on SO are called Identicons, you could use a variant of that.
See this question for more information:
http://stackoverflow.com/questions/101918/how-is-the-user-id-image-generated-on-so
You could try generating a face-like image. People are fairly good at distinguishing different face-like images. For example, see Wavatars, which are goofy face-like images in which the shape is varied, the color, the background the eyes, the mouth, and so on, leading to a very large number of different, fairly recognizable images.
If not faces, then pick some other class of fairly recognizable objects that can vary on several axes, like cars or animals or something.
See, for example, this thread on Meta about automatically generating unicorn avatars.