[Seeking advice on your art pipeline? In this article, originally published in Game Developer magazine in 2008, Bungie's Steve Theodore delves deep into the issues behind taking a piece of art from 2D concept sketch to finished in-game model, with the least waste and complication.]

You could build a nice cozy two-bedroom bungalow out of all the verbiage the games business has expended on "the content pipeline" over the years. The term "pipeline" makes it all sound very rational and linear, the sort of high-tech industrial process that involves hard hats and jumpsuited henchmen.

But let's be honest. If what artists do fits into any pipeline, it's one of those that you find in Dr. Seuss books, full of crazy loopbacks, recirculations, and about-faces.

Partly this is because even the simple part of the pipeline -- the software that's supposed to get the art out of your content tools and into the game -- is mutating and morphing constantly as game designs and engines evolved. But mostly it's due to us. Even the part of this mythical "pipeline" that we control, the concept and design side, is iterative, messy, and nonlinear. We're artists. We aren't built for assembly line production.

If you don't face that fact (and build your process and schedules around it), you'll ship stuff you hate, full of bugs you know you could have avoided, or full of crappy content that should never have left the building.

On the other hand, once you accept that the art process is fundamentally not linear, you can make some common-sense adjustments that will help make the whole crazy business a little more manageable. So this month let's look at some of the ways a concept can get jammed in the pipeline, and some of the virtual Drano you can use to muck out that old pipeline and get things moving.

Down the Tubes

If you still subscribe to the myth of the pipeline, you know that creating concepts is the key to production. It seems logical, after all, that you should try to work out all the messy creative issues in sketches, where they can be tackled quickly and cheaply.

A concept artist and an art director can spin up lots of images quickly as they grope for the elusive soul of the new character. They can also use those images to sell the new concept to the rest of the team, getting feedback from other departments to head off potential problems.

And of course -- ah, sweet naivety! -- since the character has been carefully defined in the concept stage, turning it over to modeling and thence to animation is just a matter of execution.

You can see why this is an appealing idea. Strong, thorough concept work is undoubtedly a Good Thing. It keeps the whole production cycle attuned to clear vision and goals.

It allows you to iterate cheaply and minimize risk. It also has the seductive side effect of centralizing the creative work, meaning the team can get by with less experienced artists on the production line.

There's nothing particularly innovative about any of these observations, of course; they are the stuff of many a GDC talk. They're also music to a publisher's ears since it helps them pretend that their teams have a master plan that can be followed rationally, step by step and milestone by milestone.

Unfortunately, "appealing" and "likely" are two very different things.

Pipe Cleaners

To illustrate some of the ways in which this tidy model breaks down, let's look at a concrete example.

The HiveHound (see Figure 1) was a typical game industry monster, an acid-spitting cross between a scorpion and a Doberman of the sort you meet all the time in our business.

Despite a very thorough, meticulous design process he was blind sided by the unpredictable, infuriating realities of going from concept art to game asset. So he can serve as a cautionary tale about what you can and can't get on the strength of good illustrations.

Like most characters, the HiveHound was born as a handful of bullet points in a design doc. The spec called for a medium-sized creature that operates in packs, was basically feral but had some rudimentary intelligence, and shared a kind of hive-mentality.

The drawing stage kicked off with lots of pictures from the web, then silhouettes and sketchy thumbnails.

Figure 1: An early sketch of the HiveHound, as the basic formula began to take shape.

As often happens, the gameplay and the visual concepts ping-ponged back and forth, as fresh ideas emerged and fed off one another, so that the spec which emerged from the first round of meetings was somewhat different from the original design concept.

Nevertheless, at this early stage the process was working pretty much going according to plan: lots of quickie pencil sketches and Photoshop iterations that gradually honed those vague designer commandments into something like a graphic formula for the character. Thankfully, it was all done quickly, without a lot of polish wasted on unfinished ideas (see Figure 2).

Figure 2: So far, so good -- the basic design in place without too much wasted work.

The next task was to get from "what are we making?" to "what does it look like?" Hence, another round of concept art, this time farmed out to the consummate pros at Massive Black, an outsourced art studio. The sketches coming back from the concept team started taking longer and longer and were becoming more detailed as the formula was refined.

As the visual vocabulary for the character solidified, we tried hard to anticipate future problems and head them off. Will that arrangement of neck plates be flexible enough? Can he climb a tree with those limbs? How does that ranged attack work?

Each new question sent additional drawings flying back and forth as we tried to settle those questions early, and, it was hoped, cheaply. At this point we felt like we were just about ready to really start flowing through that pipeline like $100-per-barrel crude.

Pipe Dreams

Unfortunately, the progress we were making wasn't as solid as it seemed. Events would soon show that for every problem that was sidestepped in this phase, another two still lurked behind it.

The feeling that all those problems could be spotted and solved in the concept phase was doubly dangerous. Not only did it turn the concept drawings into something more like a mechanical drafting exercise, it also created a false sense of security. Having avoided (it was felt) so many pitfalls in advance, it must surely be safe to start driving for the final layer of polish.

With a fat folder of pretty pictures in place, it was time to turn the vision into reality. Since one of the HiveHound's jobs was to star in a prototype vertical slice, he had to be a fully "next-gen" character with all the normal-mapped goodness, high poly counts, and crazy shader effects you need to impress a potential publisher.

This meant that it was necessary to create the scaffolding that would hold up this mighty next-gen edifice as well as the art itself. A mini pipeline for assembling a complex, multimillion poly version of the creature had to be put in place as part of the test, and also a shader system that could handle the glowing-acidy-lava-tube thingies that were a key part of the design.

Some of this was unavoidable investment. Unfortunately, the need to proof the pipelines translated into even more pressure to treat the design as finished. The zillion-poly version had to be built because the zillion-poly pipeline needed to be tested. The fancy shaders needed to be tweaked because the shader engine needed to be put through its paces.

The unhappy result was that the HiveHound reached the animation and rigging stage as a completely modeled and shaded mesh with weeks of work sunk into the high-resolution model, shaders, UVs, and texture work. It looked great, but serious problems were about to pounce on us.

Clogged

In a traditional concept pipeline, the evolution of the character is over by the time the design is handed off to the modelers. In reality this isn't the end of the design phase -- it's the beginning of the most important and difficult stage in a character's evolution.

A character is not just a drawing, or a model of a drawing. It's a living entity. Behavior is just as important as graphic identity for selling the character. Imagine combining the body of Shaquille O'Neal with the moves of Richard Simmons. You'd get -- let's just say, a very different character than either a standard NBA player or a standard exercise guru. All too often we treat the design of a character's behaviors as an afterthought, as a detail to be dealt with only after the "real" work of finding the character's look is done.

Figure 3: From concept to fancy next gen model. Unfortunately, the design was flawed -- the character's rear legs didn't allow enough freedom to run and gallop.

This doesn't help the artistic integrity of our characters, and it's a big risk in production terms. The case of the HiveHound illustrates both risks all too well. When the beautifully shaded and rendered character (see Figure 3) was prepped for animation, we discovered a critical flaw that had gone completely undetected through all of the concept iterations that came before it: He couldn't move right.

His proportions were loosely modeled on those of a wolf or hunting dog, but his leg arrangements had been made more angular and simpler to reinforce the insect side of his personality. Without the compressibility of canine rear legs, his fastest gait was sort of a baboon-like lope instead of the bounding wolfish run he'd been intended for.

An equally damning problem emerged when it came time to prototype the creature's ranged attacks. A lot of energy had been expended on brainstorming how the throwing mechanism would work, right down to detailed sketches of flexible "fingers" that let him flourish his throwing spikes with theatrical menace (see Figure 4).

Figure 4: Trying to head off problems by detailed concept work is a good idea. Too bad it only goes so far.

Unfortunately the concept didn't make clear that, being attached to the flexing spine of a fast moving creature, the throwing arms would gyrate wildly in motion, rendering a convincing throw animation almost impossible.

Fixing the fully detailed and textured model to allow for more appropriate animations was (it needs hardly be said) a costly boondoggle that involved throwing away a lot of otherwise good work. It also played havoc with the character's schedule. Of course, there's nothing unique in this little tale. Indeed, that's why it's worth recounting.

The kinds of problems that emerged as the HiveHound passed from concept to game asset are typical of the nasty little surprises that lie in wait for most would-be game characters (Figure 5, below, shows the finished concept piece).

Movement, gameplay, shader interactions -- there are innumerable ways in which a character can confound your plans. The HiveHound's thorough concept phase probably did head off many potential problems before they transpired, but it didn't and couldn't catch them all. It's naive to expect a concept team to magically anticipate every potential gotcha.

The problems are too varied and the expertise necessary to anticipate them belongs to too many people -- animators and riggers, designers, engineers, as well as concept artists and modelers. There is no way to plan it all out on paper.

Figure 5: The end product of the concept stage was compelling. Too bad it wasn't practical.

Light, Sweet, Crude

The only safe way to avoid long and costly detours into untenable concepts is to be absolutely rigorous about not over-committing to any design feature until it has actually passed through the whole process from initial sketch to animated character.

The concept phase shouldn't end when a sketch or model sheet is passed from the concept team to the modelers. It's not a handoff. It's taking the concept development into a new stage with its fundamental goals -- fast iteration and exploration -- intact. Turn the sketch into a 3D proportion study, and the study into a set of test animations, before anybody from concept side to the animation staff starts talking about "polish."

Treat the early models and rigs as tools for refining and proving the concept, not early jumpstarts on the grind of production. Even rapid prototyping can't completely eliminate glitches, but it can help to insure that they are fewer in number and that they are less costly in terms of work wasted and time lost.

Involving modelers, animators, riggers, and designers in the concept phase has broader implications. Expanding the concept effort can be frustrating if it leads to endless meetings or design by committee.

On the other hand, enlisting the expertise of all the different disciplines can also be a huge plus. Not only does it help you anticipate problems, it also democratizes the sense of ownership and creative input that makes great work possible. We started off with the notion that pipelines start with artists, and that's always the most important fact to remember about what we do.