As an artist, I have a huge passion for the process of pre-production; taking an idea and iterating until you achieve the correct result.
I often talk to my students about the idea of happy accidents, and the use of mood boards and thumbnails to help plan their work.
This video at SIGGRAPH really caught my attention—a group of students from Stanford University have developed a probabilistic model for component-based shape synthesis. The concept is demonstrated quite clearly in the video, but in layman's terms, by providing the system with a small range of models, it is then able to generate a far larger group of models comprised of determinable components from the originals.
If you're feeling brave, you can take a look at the students' accompanying paper. Now this takes iteration to a whole new level.