A generative art piece I really like is Dreamline by Leonardo Solaas, shown below.
Dreamline was open to the public from 2005 until a few years ago; users could enter words from dreams/of dreams they would like to have, and then, the system searched the Web for related images and conjured them into an ambiguous painting. The work was developed in both Java and PHP; PHP script made the Google Image search from the word input, and Java received the images and placed them. No complete images are shown – the work is made up of 1500 pixels in constant motion.
I think this work is a strong representation of Solaa’s fascination with the subconscious: it generates an impermanent, moving image (just like a dream). I admire that he was able to harness minimal user interaction (entering words) to still create something vividly complex and individual. He was probably inspired by other systems that follow rules, but still maintain flexibility, like the drawing machines of Jean Tinguely.
I do wish that he would reactivate the art piece because he said he had to stop the project due to outdated technology. I also believe this work would have benefited form the ability to save individual images and categorize them in some way, though this deters slightly from the transient element.
http://www.digicult.it/digimag/issue-051/towards-the-edge-of-chaos-an-interview-with-leonardo-solaas/
http://solaas.com.ar/dreamlines/about.htm