We Make You Think

WoW’s Character Model Update: A Primer

update-primer

Want to make me cringe? Make a post on Blizzard’s official World of Warcraft forums, or a fansite’s WoW forums, and demand that Blizzard updates the character models for WoW. Say that updating the models is easy and the artists are lazy. Be sure to mention how other games like RIFT, Guild Wars 2, or The Last of Us have much better looking character models. Then, include a wish-list of things you want to see on the updated character models.

I cringe because nearly everyone demonstrates a complete lack of understanding of how WoW’s models work, let alone character modeling at large. Updating the models is a monumental task. There is a lot going on under the hood. There isn’t some magic button the artists can click that instantly upgrades everything to a higher polygon count. (Even if there were, there’d still be quite a bit of work to do to get the models ready for use in-game!)

Updating character models isn’t going to be an easy task. First, let me try to explain what a model is. We can get into updating the character models later.

Models are made out of polygons. Most game engines render three-sided polygons, which anyone can tell you are called triangles. Some game engines are capable of rendering four-sided polygons—quadrilaterals (or quads, for short). Most modeling packages are capable of creating quads, and this is what 3D artists often work with.

Artists prefer working with quads because subdividing quads produces a much cleaner result than subdividing triangles. A quad will subdivide into four smaller, uniformly sized quads, while triangles are subdivided into three quads. A subdivided triangle creates what artists call “poles” which create unsightly bumps when the model is subdivided further. Poles can also cause textures to become stretched, which can create artifacts later when texture maps are baked.

In a photorealistic game, such as Uncharted, character artists will first sculpt a character mesh. If you want to be pedantic, a concept artist will create several iterations first, those concepts will be used to create a character sheet, and then the character artist will use that character sheet as a reference when they’re sculpting. Sculpting is like working with digital clay; artists can squeeze the model, extrude it, bend it, and use tools to shape it. Every wrinkle, skin pore, and scar is created as part of the 3D model. Artists will often use their own brushes to create the look they want. However, creating this kind of detail requires a model that has hundreds of thousands of polygons.

That’s where subdivision comes in. The artist will start with a low resolution mesh. When talking about 3D models, resolution refers to the number of polygons; a “low resolution” mesh would have a polygon count of roughly 10,000 (or less!), while a “high resolution” mesh is usually in the ballpark of 250,000. The artist might choose to start with a generic base mesh—a mesh that looks sort of like a male or female, but is devoid of any detail. They may opt to start with a low resolution sphere or cylinder in Zbrush and sculpt the basic form. In either case, they’re starting with the big shapes and defining the silhouette before subdividing and working on (slightly) smaller details like musculature, wrinkles, and eventually pores.

These high-resolution models are unusable by game engines. To get around this, artists will bake details into a texture map. If they painted on the model—the way you might paint your Warhammer miniatures—they can bake that paint into a diffuse map. They can bake the wrinkles and pores into normal maps—the normal is the direction a polygon is facing. They can even bake shadows and other lighting information into a texture so that the game engine has less data to calculate at runtime.

Once all of the high-resolution model’s details have been baked into textures, the high-res model is still unusable by the game engine. What artists need to do is recreate it at a lower resolution. There isn’t an exact ratio that says a 250,000 poly model will be reduced into a 12,000 poly model; it depends on the needs of the project and the skill of the artist. If the project is aiming to invoke feelings of nostalgia, the low-resolution model might be blocky and in the range of 2,000 polygons. Photorealistic games often use higher poly counts.

The process of creating a low-resolution model—or rather, recreating the high poly model at a lower resolution—is called retopology. This would make sense to you if I had explained what topology is.

In geography, topography refers to how the land is shaped. It’s not just elevation and contour lines; it’s the mountains, the plains, the waterfalls, and the crags. The Great Plains of North America have very boring topography, which surprises no one on account of the endless expanse of flat grassland. By contrast, New Zealand has very interesting topography, which is also not surprising because it was formed by millennia of volcanic activity. I mean, you’ve seen Lord of the Rings. That was all shot in New Zealand. It’s a fantastic place.

newzealand

I could have watched the entire Lord of the Rings trilogy and gotten a screencap of faster than it took me to find a usable photo in the public domain. Not a *decent* photo, mind you—just a usable one.

I’m not talking about topography. Topology refers to how the polygons are laid out. It refers to how the edges flow through the model and how they loop around. It also refers to the form and the detail. A good low resolution model will capture the silhouette of the model, but not feature very much detail—after all, that’s what the various texture maps were baked for. Topology also refers to how well it can deform.

Consider a sheet of paper lying on a table; the paper is flat and has four sides. I can create a sheet of paper with a single quad! That’s good topology. However, if I need this sheet of paper to deform—perhaps it needs to tumble in the wind or the edges need to curl as it burns—then it has bad topology. The paper does not have the edges needed to allow it to bend.

Simply put, retopology is the process of simplifying a mesh by laying out the polygons manually. Capturing the form is more important than recreating the detail. A deformable model is more valuable than a static model. Retopology is, literally, redoing the model’s topology so it is a usable game asset.

Of course, this is just the pipeline for a photorealistic game. World of Warcraft is not a photorealistic game.

Clearly.

Clearly.

World of Warcraft uses a very stylized aesthetic. However, this does not preclude it from the standard modeling practices. Stylized games typically use hand-painted diffuse textures, so there is very little reason to do a high-poly sculpt first. (This isn’t to say that high-poly sculpts aren’t done for WoW. There absolutely are, but right now, it seems to be mostly done by texture artists to create a base texture to paint over.)

However, World of Warcraft is still a 3D game. Blizzard’s artists still have to create usable models. A stylized aesthetic does not mean artists can ignore basic principles of topology. Organic shapes cannot be a grid of polygons; they need to be round. The models still need to be able to deform.

Spiderman?

Spiderman? (Ullyr by daren)

Organic shapes are typically round. Most characters are organic, unless it’s a robot, but that never happens. Therefore, a good character model is going to be round. So in the image above, both models have a round shape, but the one on the right has good topology. This is because there are lines—edge loops—that go around the eyes and mouth. This allows the model to deform realistically when it’s being animated. The rigid grid layout on the left can work, but it’s not optimal and the technical artist rigging that model will strangle the character artist in their sleep.

Edge loops are laid out to mimic facial muscles. When the mouth opens, the cheeks will stretch and wrinkle like you’d expect them to. Raising the edge loop over the eyes will recreate the appearance of the outer frontalis muscle contracting. That sentence should have made you raise an eyebrow—literally. That’s the muscle responsible for arching your eyebrows.

Good topology allows for clean deformation, but it doesn’t actually deform the model. Animation handles the deformation, and skeletons are responsible for animation.

No, not like that.

No, not like that.

If you were to cut a character in half, you would find they’re hollow inside. Game characters are as boneless as chicken nuggets. What animators use to bring a model to life is called an armature or rig.

An adult human body has approximately 206 bones in it. In AAA games, a human character will have something like 100-125 bones; again, the needs of the project will dictate how many bones there are.

Consider an arm, going from the shoulder to the fingers. In a real human, there is the clavicle, the scapula, the humerus, the ulna, the radius, the four carpals in the wrist, and 19 bones in the hand; a total of 28 bones. In a game, that would be reduced to an upper arm, forearm, and 15 (or 16) bones in the hand. There wouldn’t be any bones in the shoulder or the wrist, because those aren’t needed. The spine, of which humans have 26 vertebrae, is typically reduced to 4 bones: pelvis, lower back, upper back, and neck. A human leg will have 30 bones in it, but a game character’s leg will have four.

Aside from the difference in number, real humans and their virtual counterparts have similar skeletal structures. The only bones omitted are those used mainly for support—such as the ribs—rather than locomotion. In this way, armatures are like our own skeletons; there are bones connected by joints. However, the skeleton analogy starts to break down when we get to the head.

For meat sacks like us, our eyebrows are controlled by a single facial muscle, the frontalis. The eyebrows for game characters are controlled with two bones each. Bones in an armature are points of articulation; they can represent bones (skeletal) or muscles. An armature is just a framework for animation, after all!

There’s a lot more to animation than what I’m explaining. I haven’t even covered keyframes, nor explored the discrepancies between why Worgen have 205 animations while every other race has 196 animations. But that’s a topic to discuss another day.

Onto the good part: In World of Warcraft, the Pandaren have good topology. They had better, considering their models boast an impressive 6,200 triangles! They also have about 125 bones in their rig. Half of the triangles and 50 of the bones are used just in the face, which is what makes the Pandaren the most expressive race in the game. They can smile and frown; their cheeks can squash and stretch; their brows can furrow and lift.

In fact, Pandaren faces resemble faces from other AAA games. Here’s a side-by-side comparison with Nathan Drake from Uncharted 2:

uncharted_comp

For comparison, every other player race has between 30-50 bones for the entire model. The number of triangles ranges from 800 (Gnomes) to 2,600 (Worgen). The Pandaren are leaps and bounds ahead of the older races.

And that’s just a brief general primer about what a character model is, but it’s scratching the tip of the iceberg. There’s a bit more to what Blizzard does to make a model. Check back on Thursday when I explore Blizzard’s M2 model and why that makes it a monumental task.

You May Also Like:

No comments

Trackbacks/Pingbacks

  1. WoW's Character Model Update: A Closer Look | Clever Musings - […] the old character models is a monumental task. Last time, I covered the basics of what a model is. …

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

© 2011-2017 Clever Musings All Rights Reserved