Texture Targeting

I made it to the San Francisco Postmortem, a get-together for bay area game developers, for the first time last Tuesday. I finally got to meet the creators of Natural Selection as well as reconnect with some friends from college, Sam Hart and Chris DeLeon. The guys at Unknown Worlds are busy working on a sequel to Natural Selection on a shoestring budget with minimal staff, and they are doing an amazing job of doing things as efficiently as possible. The biggest bottleneck though for a modern game is the incredible quantity of art assets required to achieve the graphical standards gamers expect. I've been interested in texture synthesis since I took a graphics class in college, and decided this would be a good opportunity to experiment with it in the hopes that it could speed up creation of texture sets. I've started a texture synthesis project on Google Code, and there are two bare-bones utilities there that are ready to start playing with. The code is all based on techniques from Efros and Leung with a variety of experimental modifications of my own. The ultimate aim of this is to produce something akin to an "automated super clone tool" such that an artist can make a few textures for a theme in exacting detail, and then synthesize the rest from sketches using this software.

So far I haven't been able to get satisfying results for any non-stochastic textures, but that's my aim. With this technique there seems to be a dramatic trade-off between giving the artist any control at all over the resulting texture, and the amount of detail present in that texture. If anyone has an interesting results or ideas in that regard, drop me a line.