Shamus over at Twenty Sided has a great series of articles up on FUEL, specifically the game design aspect of procedural content, and how it fosters exploration in the FUEL world.
It’s well worth reading through all of them if you have any interest at all in how games are made.
Procedural content is a fantastic tool for devs. It multiplies art assets and labor, and makes higher level art direction easier in some ways. WoW made extensive use of procedural terrain generation, and I suspect pretty much any MMO has used at least a simple fractal generator or three, at least as a starting point for the huge swaths of land they tend to boast. Crafting all of that by hand just isn’t feasible. Hand-polishing places here and there, and lovingly crafting certain high traffic locations is smart, to be sure, but you can do a LOT with clever procedural algorithms.
This is one good example of fractals that show up when you know what to look for (thanks to Postcards From Azeroth):
The edge on that lake is very fractal. Compare it to the Mandelbrot set fractal over here. You’ll find this sort of thing all over the place when you look at large scale geography in games. Fractals tend to look natural enough to fool the eye, at least as long as you don’t look too deeply. There are entire programs built around this. Bryce is perhaps the most common one, but Terragen looks to be one of the more powerful ones. (I keep meaning to experiment with it… when I have all that free time, right?)
Whether you use one of those programs or an in-house terrain generator, the principle is the same, though. Let the computer do the mindless, repetitive vertex shuffling according to some simple rules that can generate complex output. (Pretty much what fractals are all about.) Computers love to do that sort of repetitive, mindless number crunching, while computer artists and engineers tend to see it as a waste of time. (Not to mention mindless and morale-killing.)
It’s a bit like brushes in Photoshop, or filters. They exist as visual “macros”, quick and easy ways to perform repetitive tasks, hopefully freeing up expensive artist time to work on art direction and specific set pieces. That’s a significant part of why Photoshop and Painter even exist. I could reproduce the Mona Lisa pixel by pixel in MSPaint, but why in the world would I want to beat myself up that way? Computers waste a lot of our time already, why not let them shoulder the load when they are good at it?
I worked with Renderman for a while in college. In one class, I wrote shaders from the ground up, effectively creating textures for 3D objects completely in code. There are some shots of my work over here:
Renderman (There are also some cool eye shaders over there that I had a blast putting together. The mesh for the eye was just a simple NURBS sphere; all of the subtleties of the cornea and coloring came entirely from code.)
OK, it was done in Renderman code, which is more like scripting than coding. Still, there was no painting involved, no 2D graphics program, just code and a 3D mesh from Maya.
Procedural Gourd Texture
Everything that made this gourd’s coloring was typed up in a text editor. The assignment was to take a real gourd and replicate it with a simple Maya 3D mesh and the Renderman code. The color, texture, lighting, surface lighting properties and even slight bump mapping (I don’t think I used displacement mapping on that one) on that gourd are entirely text-based. If I, a paltry college student, could do that, imagine what the Pixar guys get away with.
You can do a lot with code.
That might sound a little odd, coming from me, an artist. I’m one of those newfangled artists, though, who sees great value in this sort of thing. I loved the “tree painting” tool I used for EA. It let me “paint” where trees should grow, how dense and how tall, on a Maya surface, and it would bring in preconstructed trees and scale them appropriately. I effectively painted forests onto the world. It really could have gone a step further and had the trees themselves be procedurally generated. Trees are perfect for that sort of thing.
Roads are another natural fit, considering they tend to either fit a grid or flow with the terrain (itself procedurally generated). Sure, you need more constraints in how they actually behave so that they make sense, but it’s precisely that sort of process that game programming deals with all the time. It’s all in finding the simplest rules possible that will, when given a starting point and a bit of randomness, can produce something that looks natural.
In a way, it’s telling the computer how to process rules like we teach art students. If you can give the computer the code equivalent of the rules and principles of art, like balance, rhythm, line, proportion, variation, motion, color theory and so on, it can generate a lot of art in short order. (All that artists do is apply the same rules to whatever ideas rattle around in our head.) It may not have the spark of creativity that makes the finest hand-crafted pieces as great as they are, or the rationalization that allows us to think a Pollock painting is high art, but computers, sufficiently educated, can do a lot.
Though, it should be noted that educating them takes time and effort, as well as an artistic eye. Computers are also notoriously stupid (or exceptionally smart) in that they are rigidly bound to internal logic and thinking “inside the box”. They are fantastic tools, but must be used the proper way to be of any use. You can’t just put Deep Blue in a room and tell it to whip up the perfect video game. You can, however, use them to shoulder a lot of the inane repetition in the development process.
Shamus describes it in depth over at his place, so again, I heartily recommend his articles. I’m mostly just chiming in as an artist, noting that I love it when the computer can help make my job faster and easier. I suppose that at some point, it could be argued that a sufficiently advanced computer could do the job on its own, but I’m confident we’re at least a few years from that situation. And, well… once that day comes, I’ll be telling the computer how to do the job I once did, so things will work out.
Tish Tosh Tesh 
Was it you that recently discussed being funneled from one zone to the next in MMOs in order to reduce the CPU load when your processor needs to render the new scenery and environment change?
I remembered that article earlier this week when I went from Elwynn Forest into Redridge and it started raining. It wasn’t like it went from not raining to pouring down in the space of one step, it was noticeable that the weather was fine there while it’s raining here, but moving from one zone to the next was not like running into a wall of water. When I stopped and looked back I also couldn’t really pin down the point at which it went from “raining” to “not raining”.
If you cross the river from Elwynn Forest or Westfall into Duskwood and it’s raining in one of those zones but not the other, you’ll notice that partway across the river it starts raining. Clearly the river is being used to easily separate the two zones without visually jarring the user with a noticeable scenery change. But even the change between Elwynn and Redridge is hard to discern, at least to my casual eye (it’s probably more visible to someone with artist-trained eyes like your own).
Even when the weather was fine in both zones and I stopped and tried to pick out the border I found it impossible to say “there it is”. I could see that there was green foliage in Elwynn, with plenty of vegetation and trees, while Redridge was dry and clearly in need of a good rain, but I couldn’t really say where Elwynn ended and Redridge began. There must be a border between the two zones – else how does the game know when to change your location from being in Elwynn Forest to Redridge – but I just couldn’t find it.
I remember that article, but I didn’t write it. I’m not sure who did. 😦
Zone changes are indeed a great place for hand-tailoring the experience, while the internal zone content can be largely procedural. Savvy art direction can indeed smooth out those boundaries, and it’s a key to crafting a sense that the game world is an organic whole, not just a series of rooms.
Though looking at a bird’s eye map of the whole game like this one at WoWMap.com shows that those blurry boundary zones are actually pretty small. Big enough to fool the eye of the player at ground level, but when viewed on a higher level, clear blobs of fairly sharply delineated “climates” exist. That map viewer is actually a great tool for picking out some of those high level fractal designs and several other terrain generation shortcuts that you’d not note on the ground.