AARON’s new colour system uses three lists of numbers, one for hue created from an additive series with a random start, one for saturation generated randomly and sorted into low/medium/high ranges, and another for value created the same way. AARON then chooses the hue according to the kind of object, and the saturation and value from the low/medium/high ranges according to the probabilities assigned to them. There’s some error correction and some mach-band-like generation of slightly lighter and darker colours for edge adornment but as Harold said it is a very much simpler system than AARON was using before.
It is also a strikingly successful system aesthetically. The colours and the contrast of the images are pleasing and interesting, sometimes mellow, sometimes dramatic, always “creative”. Perhaps this has to do with the new system matching our cognitive perceptual system’s preferences in some way.
I asked a question, badly, and got an answer with some interesting details about how AARON handles intersecting objects. A Furtherfield reviewer was there so hopefully a review will be up on Furtherfield soon. Many thanks to the excellent Computer Arts Society and Imperial College for arranging the talk.