There are several factors we have to balance in coming up with a style of animation and rendering for “Lunatics!”
You might think that 3D animators should always try for maximum realism
(“photorealism”) when making animation, but this is not necessarily a
good idea. First of all, the human eye is extremely good at spotting
errors in photorealistic renderings and especially in animation.
This is the basis of the problem known as the “uncanny valley effect”:
if you have extremely photo-accurate models and renderings of
characters, then even the slightest error in movement creates a
disturbing “creepy” effect. Such animations are often described as
“zombie-like” or “doll-like”. This is because we are very sensitive to
tiny differences in the way real people move.
Tim Farmer, with simple previsualization materials and
normal photorealistic shading. This makes him look more like a real
object, but maybe not more like a character. Model by Bela Szabo, design
by Daniel Fu. However, just because the design is a little more
stylized, this model is already an improvement over a photorealistic
character.
Cartoons get around this problem by having
characters which are highly stylized. Your brain fills in the gap
between what you see and what is intended. That is, you’re cooperating
in the suspension of disbelief required to enjoy animated films. Since
you’re already interpreting non-realistic images, it’s not disturbing
when they don’t move in exactly realistic ways.
So to make our characters more approachable, we need to give them a little bit more of a “cartoon” look.
On the other hand, since we want to have a level of realism in “Lunatics!”
in regard to space technology and settings, we’d like to be able to use
normal proportions. Highly distorted cartoon characters can make that
hard — sets and especially props have to be adjusted significantly to
fit the characters.
We compromised on this, but our characters are very close to real proportions.
So we’re seeking other ways to reduce the “uncanny valley” issue. One of these
is to use “non-photorealistic” (NPR) rendering. This tends to make the characters
look as if they have been drawn or painted rather than photographed. As with other cartoon
techniques, this tends to make your perception a little more forgiving
about movement.
This materials test render shows the model of “Georgiana
Lerner” with toon-shaded materials, but no lines. This is how Georgiana
is rendered in the teaser demo that we started our video with. This
avoids some of the creepy “doll-like” look, but it can look a little
dull or poorly defined since the shading is simplified so much.
One thing we’re still considering is whether the character would look
better with Freestyle line-drawing enabled. This can give a toon-shaded
character a bit more definition — the result can be a bit like an anime
character. Here’s a fairly subtle version of this:
The same Georgiana Lerner model, but with some thin
outlining provided by Freestyle. This tends to give her a bit more
definition, and makes her look a little more cartoonish.
There are a lot of parameters to adjust with Freestyle, though, which can
alter the result dramatically. It’s also a pretty time-consuming
rendering step. We can do it if we want to, but it will increase
rendering times.
So far, I have preferred the unlined rendering
to the lined versions I have tested, but I want to continue
experimenting with this before making a final decision. It may be a
matter of getting the line styles just right. Most likely, we would use
this technique on characters, but not on backgrounds — somewhat
mimicking the style of Japanese anime, where backgrounds are painted,
while foreground characters and objects are inked and colored.
Of course, this is one of the last directing decisions I need to make —
as it only affects the final rendering and compositing stages for the
project.
