Artsy Rants

The Anatomy of a 3D Character Build: How I Create a Virtual Character

  • January 31, 2016

I’ve already posted about the fun 3D gargoyle character I’m creating but today I decided I’d like to go more in depth into the process of getting the model from concept to fully-realized and ready to animate. That last is still a work in progress (a “WIP” in the vernacular) but it’s quite close to completion now. I’m always excited when a build gets to this point because then the real artsy fun can begin!

To recap, I became intrigued by a gargoyle statue in my neighbor’s coffee shop and decided I wanted to do something related as a 3D character. This is a photo of the statue itself:

Gargoyle Statue

Dude: you know you want to buy me a coffee.

With that image in mind I set about sculpting the gargoyle’s form. I won’t go into the gory details of the build process here because I’ve already posted about how I do that sort of work. You can see an example of a torso build in my post about working in virtual clay as well as watch a video of me sculpting a very basic toon style head.

Two challenges of this build were the “digitigrade” style legs and the dragon style wings. Although I’ve built many character models, even a 3D cartoon bear, all of those models have featured “plantigrade” or human style legs and none have had wings. I tried, and failed, to craft pleasing legs from my mind’s eye alone and eventually realized something all good modelers should know: work from reference material whenever you can. A quick session with Teh Google and I had a nice collection of photos of various animals in fleshed and skeletal form. I also found this extremely useful sketch related to “anthro” or anthropomorphic animal characters (said leg style being exactly what I was trying to get to):

leg reference image

Who knew there were so many variations?

The wings, once I thought about it, turned out to be rather simple to create. They are just arms, hands, and fingers that are proportioned so the arms are short and the fingers fanned out, long, and slender. The only difference was adding the membrane but thanks to the miracle of modern modeling software that was a snap. One note was needing to make sure the membrane’s polygons were aligned with the finger joints to insure smooth deformation later on.

I banged out the primary build in roughly two and a half days—not working full-bore on the project, but putting in a respectable number of hours per day. Again, the unfamiliarity of the leg/hoof style plus experimenting with several approaches to the wings made the build a little slower than normal for me.

This is a screen shot of the full model in low-poly mode (the version of the model that I actually worked with):

wires cage

Pretty ugly and blocky, no?

This is the same model in subdivision surface mode, where the blocky-looking polygon cage is automagically smoothed (this is, of course, the version you actually see in renders):

wires subd

Much purtier!

A staple of model construction at this point is the “turntable” render where you can examine the model from multiple angles under typical studio lighting. It will reveal any problem areas in the basic geometry that could cause rendering artifacts or unpleasing silhouettes. You can also get a start on the surface materials you want to use. Here is one such test for this model using a surface I created in homage to the silvered one of the original statue:

Your basic Power Twirl

Note that although the model is not yet rigged at this stage I partially posed the wings to get a better look at them. Of course the version for rigging has them aligned to the major axes to make adding bones easier.

Then comes the rigging! The rigging process is possibly the most tedious one that exists in the modeling/animation world (next to waiting for a lengthy render to complete). The current state of the art has come a long way in that regard, thankfully, and I have gleefully embraced every advance in rigging technology as it’s become available to smaller Studios like mine. Nowadays I can (minus testing and tweaking) bang out a fully-realized body rig in just a couple of hours.

Standard practice is to start a rig with a “root” bone that will be the One Bone to Rule Them All. Moving this bone around will take the entire rig with it. For bipeds the best place for this root is in the hip region. From there individual preferences vary; mine is to build down to the ground first, getting the lower body and legs in good form and then building from there up towards the arms and finally the head. A lot of folks like to get the spine in place right away but I march to the beat of a different drummer.

Here’s a look at the lower body portion of the gargoyle rig. The tail assembly is very heavy, 30 bones, because I want the tail to be extremely fluid when I start animating it. I tried a couple of lower-count versions and was just not happy with the segmentation under tight curves. Fortunately my Mac Pro has the horsepower to handle this sort of thing with no issues.

rig legs

Dem Bones

Then it’s a slog, adding bones and building up the skeletal structure and the associated control objects for posing it. Long gone are the days of directly clicking bones (or boxes in a schematic of the rig) to pose them. Those control methods still exist but I’m not Old Skool enough to want to use them.

Now I’ll go into three views of the “final” (no rig is ever final, even if you had to deliver it to Production. There is always something that could have been done better) rig for my gargoyle as it stands today. This first one is what I’ll be looking at when posing and animating the character. Note I changed the texture of the model to a material that looks as though the gargoyle were carved from stone—the traditional medium for this character.

gargoyle rig 1

Which ones are the twerking bones?

Here’s a view of the same rig with the model hidden so you can appreciate unobscured the elegant beauty of a bunch of virtual bones.

skeleton rig 2

Look Ma: no skin!

So, what’s up with all that orange blocky stuff? Those are called “cage deformers” and they are one of the newer fun toys in the animator’s arsenal. In practice, no matter how good your model and rig, the lack of actual muscles under the skin keeping the skin pumped up can cause the model surface to lose volume when deformed. Shoulders, elbows, knees…the typical problem areas are legion. There exist many ways to fix such things, such as hold bones, “muscle” bones, corrective morphs, and the like. Deform cages yield the same result: the ability to tweak geometry in a pose to correct deformation problems. The difference is that, IMNSDHO, being able simply to drag cage points around to tweak geometry on the fly is a vastly easier method to apply corrections. Since the cages also deform with the bones, and the fine-tuning you do by dragging the cage points around gets saved as key frames in your animation, you can pose and tweak on the fly with scarcely a pause. Cool stuff!

Here’s one more look at the skeleton with the cages faded out so you can see more detail of the actual bone structure:

skeleton no cage

Gargoyle Uncaged

As you may have guessed, all those colorful shapes are the controllers for moving the rig parts around. Creating such controllers has long been both extremely tedious (IMNSDHO) and a black art where every rigger has their favorite control methods for various parts. That situation has changed tremendously in recent years, as “auto-rig” plugins proliferate. Such tools typically offer both fully-realized standard rigs for a variety of creature types and a library of rig parts you can toss in ad hoc when constructing something that needs a custom treatment. I am rather fast at dropping and linking bones manually but I believe I’ve mentioned in the past how I despise hooking up and programming control items. No real reason I can point to, I just personally find that task a chore and a bore. The drop-in rig parts usually come with pre-set controllers and the better plugins even give you a choice between multiple control set-ups for each part. Good plugins also allow you to work seamlessly with the basic rigging operations giving you complete control when you need to add stuff that might not be in the library. My personal favorite rigging tool is called Rhiggit Pro, the master work of one of Lightwave3D‘s most renowned rigging experts who is also a gifted plugin coder. Rhiggit meets every one of my criteria for a production level rigging tool and the author is wonderfully responsive and helpful when I have questions.

Once your model has passed the turntable test, and you’ve checked out your rig with some test deformations (straightforward stuff like moving the limbs around into various positions, including extreme ones, and revising as needed) it’s time to do some test poses on that bad boy. I like to come up with a cool scene and place the new model therein so as to evaluate both its deformation and its appeal in a “real scene.” Partly clinical, partly just because I dig doing this stuff and playing with a new model is fun.

Here is the initial pose test for this model:

pose test 1

My Preciousssssss….

For a first pose I was pretty happy with how this image turned out. Chief issues were loss of volume around the knee joints, some oddness around the shoulders, and the hip joints weren’t deforming as I wanted them to. For the first two issues it was time to start tweaking the cage deformers to smooth out the shoulders and flesh out the knees:

test pose 2

Kneeded some work.

Ah, much better! But wait: what’s up with those hip joints? Ugly creases! Seems when I tweaked the thigh deforms I bled the polygon shifts into the hip region. No problemo, because I thoughtfully added a deform cage for the hips and this rig comes with some nice twisty-bone controllers to help unroll stressed geometry in the limbs. A little tweak here, a little tweak there, et voila:

more tweaks

Almost there….

Now we’re cooking with gas! My one critique at this point is there is a bit of segmentation along the edges of the wing membranes when viewed at this angle. Smoothing normally takes care of that but in these cases the silhouette doesn’t lie. The current geometry is at the “sweet spot” where the membranes deform beautifully when the wing digits are posed; more or fewer polygons and they start to crumple in odd ways. What to do?

Turns out this is an easy fix, at least in Lightwave (I’m certain most—if not all—other titles offer something equivalent, just being parochial here). Remember the concept of subD and how it smooths the base geometry? Well, that’s where the problem lies in this render. The “sub patch level” is configurable and in my test render it was simply set too low for that particular pose angle.

Here is the posed model if rendered at sub patch level zero:

sub patch 0

A cubist’s nightmare

Note how simple is the wing membrane geometry; that single seam allows it to bend perfectly when the digits are posed and give it that nice billowing look without any odd motions creeping in. Here is the same pose at the default level, three:

sub patch 3

Close, but no cigar.

That’s the level those test renders were taken at and, for most situations, it’s fine to leave it there. Like I said: smoothing takes care of almost all segmentation issues; it’s when a particular scene introduces a full-on foreground silhouette like the wing’s edge that the segmentation becomes apparent. So, now, here is the same pose again, but with the sub patch level cranked up to ten:

sub patch 10

But, does it go to eleven?

As you can see, the segmentation has been massively reduced. Note that in each of these cases the actual model has not been changed one tiny bit, only the level at which the render engine calculates the smoothed version. A lower value than 10 might be plenty but since this is a one-off static render there was no need to spend time trying to find the best balance of complexity and good render results.

Why not just leave it at something really high all the time? Render times, my droogs. When you’re in production, and deadlines come into play, you need every precious second available and in any 3D production (at least for those of us with modest render farms or with only a single render machine) you should assume roughly 60% of your timeline will be spent rendering. You need to fine tune your scene so as to use the minimal amount of resources necessary to achieve the desired results (unless time is of no concern to you). For this guy I expect most scenes are going to be rendered at the default of 3 and I’ll only need to crank the level up on specific scenes where the wing edges are rather close to the camera and/or seen edge on.

Without further ado here is our hero all smoothed out and tweaked to the nines (well, the tens):

patch 10

Ready to terrorize Gotham.

Ah, now we’re talking! In Lightwave the desired sub patch level is just typed in and I have no idea what the upper limit is. Might be fun some day to find out, but not today.

And there you have it: my workflow when creating a new character model taken from conception to final posing tests. The next time I post about this cute li’l guy we’ll talk animation stuff.

Until then!

Spinland Studios, LLC is a high-tech branding and marketing studio in the Mohawk Valley of Upstate New York. We leverage the power and magic of 3D modeling and animation to take your company’s image places you can only imagine. Defy conventional marketing and bring your brand to life! Visit for more information and examples—then hire us to boost your company’s marketing image into the 21st century!

Spinland Studios, LLC is a proud member of these organizations (click each to learn more)