As you can see, this ferocious monster is a ball with two cones sticking out of it. The interesting thing about him is that just four images were used to animate the face.
There's more than one way to animate the face of a cel-shaded character in LightWave 3D.
Fighting Crush Valkyrie and Avalon both use image maps to animate the eyes and the mouths of their characters. I remember reading an article about it being a combination of Macromedia Flash, Magpie Pro, and good ol' traditional drawing skills, though I've forgotten the exact details (dammit!!!). Robert Baldwin and Sergio Rosas developed this solution to the anime-face problem, if I have my facts right.
At SIGGRAPH '98 I got to see an anime short by Ace Miles. I didn't see the actual model, but I'm 99% sure the girl character's mouth was built into the mesh, instead of image-mapped ("Not the monster from planet Gottimus!").
This page will cover one way (but definitely not the only way) of surfacing an anime character's face.
Planar image maps are the easiest to create. If you have a guideline for the proportions, you could probably even draw the facial animations by hand (then scan them in and color them in the computer).
For this animation, though, I used Adobe Photoshop. I started by setting the "face" polygons in LightWave to RGB 255, 255, 255, 100% Luminosity, and 0% Diffuse; the Light in the scene was then set to 0% Intensity. I moved the Camera back and raised the Zoom Lens setting until I got a near-orthogonal view of just the face of the oni. Then I rendered a "High Resolution" (1280x960) .TIF (without any anti-aliasing). After cropping the results in Photoshop, here's what it looked like:
I used the Pen tool to define the eye, mouth, iris, and tongue shapes. After creating non-antialiased (important!) selections with these paths, I used Edit->Fill to fill them in with color, and Edit->Stroke (Inside with a Width of 4 pixels) to create the "ink lines." I ended up with this:
The colors here are set to their "shadow" values. I created an alpha map so that I could toy with the skin tones in LightWave:
...and because I wanted the "paints" inside the mouth to never have any sort of shading (most, if not all anime characters that I've seen had no shading whatsoever on the pink tones inside their mouths), I created this Luminosity/Diffuse map:
In Layout, each of these maps plus Automatic Sizing resulted in this:
In Luminosity, I used the Luminosity/Diffuse map as is. In Diffuse, I clicked the "Negative Image" checkbox. This guarantees that the mouth area would be 100% Luminous and 0% Diffuse, so that it would always have a Zone 4 Paint Brightness. (1-12-99 Correction: I was wrong. Although it seems that the mouth falls into Zone 4, Zones don't care about an area's Luminosity. The mouth is 0% Diffuse, so it actually falls into Zone 1! Because Zone 1 is set to 100% Paint Brightness, Super Cel Shader leaves the mouth's original RGB values intact. But, because this area is also 100% Luminous, 100% of the original color is added on top of the mouth's shaded RGB values. In short, the mouth's RGB values are doubled because x + x = 2x).
I also turned off Texture Antialiasing for each of these image maps. Because I used high-resolution (1132x932), non-antialiased maps, I got crisper "ink lines" for the face than I would have if I had used low-resolution, antialiased maps. Even on these high-res maps, Texture Antialiasing blurs the lines:
(If you're wondering why the shape of the "face" polygon render doesn't quite match the shape of the face above...it's because I changed the shape of the face later on. The render still worked as a general guideline, though).
For fun, I cut up the image maps into four new images: a color map for the eyes, a color map for the mouth, an alpha map for the eyes, and an alpha/luminosity/diffuse map for the mouth:
25%-Sized Thumbnails of New Maps
I applied the eye color map first, then the mouth map. I set the texture center for the eye maps to XYZ 0, 0, 0 and used a Reference Object to raise it 6.6cm on the Y axis (the oni's body is a 1-foot diameter ball). The texture center for the mouth maps is XYZ 0,-4cm,0; it's also mapped to a Reference Null. I used Stretch on these Ref Nulls to "animate" the face:
Because the Texture Center of the eyes had been set to XYZ 0, 0, 0, the eyes stretch towards their own centers. The mouth texture is sized at XYZ 20cm, 10cm, 0 and is centered at XYZ 0, -4cm, 0, which is why the Stretch affects mostly the bottom of the mouth. At frame 5 both Reference Nulls were set to a Stretch of XYZ 1.0, 0.0, 1.0--causing both textures to "disappear." I figured no shapes were better than weird, squashed shapes (like the eyes on frame 6...hm...)
Well, it's not the best way to animate a face (this oni won't be playing the title role in Hamlet anytime soon), but for some anime characters (Ranma 1/2's oni in "Faster, Kasumi! Kill, Kill!," for example), this could work just fine. It might also work for actions where the mouth doesn't close all the way (laughing, panting, crying).
Here's a shot at "panting" (with the same image maps):
I don't know, myself. Right now I figure anything that looks good, is good.
That said, here are some ideas.
Be careful with LightWave's Camera settings. The default Zoom Lens is set to 3.2, which is a wide-angle lens that emphasizes the "3D-ness" of your models. Higher Zoom Lens settings will "flatten" the look of your models, which might improve the "2D" illusion during close-ups:
Not that wide-angle lenses and celshading don't mix:
On another note, here's one reason why you may prefer a mouth modeled into the mesh over image maps:
This doesn't necessarily tip off the audience to its 3D nature (high frame rates and extraordinarily precise movements are probably the first things people notice). I used to worry about it until I saw anime like Wedding Peach and She May Be VERY Habit-Forming!, both traditional-drawn animation which occasionally features animated mouths very similar to this one. "Cute" characters can get away with profiles like this; more "realistic"-looking characters (like the cast of Ninja Scroll) may require a mouth modeled into the mesh itself.