Major Fun with C.I.N.D.-E 1.4

Sunday, March 7, 1999 - Animating Cind-E (Part Two)

Before I show you how the animation turned out, I need to update a couple of facts.

The first is that Alias 8.2 didn't actually lose any data from my 8.0 (or 7.0) model of CindE. It was user error. (That means me.) The problem was that I distinctly recalled animating the original animation above by moving NURBS curves around. Recently when I went to pick on those curves, nothing happened, so I assumed the connection was broken.

What actually happened was that I was using the wrong "pick" command to pick the CVs on the curves. In order to do cluster animation, you have to pick with a different command. Which is lame, but what do you want for ten year old software. Which also why they wanted to invent Maya - when they say "Everything is a node!" like it is some wonderful incantation, well, it is. Out of context, it sounds about as meaningful as "All letters are in the alphabet!" or "All computer instructions are really binary!". But what it means in context is that it is easy to hook up bits of geometry to each other and even to other features like colors without all of the arbitrary limitations built into the old software.

But I digress.

The second thing that is interesting is that once I re-read the manual and learned how to pick things properly, I was able to go back and make an animation of the NURBS animation curves only - the curves that are driving the CindE animation above. It's rather amusing and fairly interesting - it looks kind of like an animated cat face. Check it out. (59K) (It doesn't include the head motion for the hair flip.)

This was interesting because as I went about the process of hooking up my rotoscoped mouth curve to the original geometry. Overall I think the original animation was too subtle - it needed a little more exaggeration - and you can kind-of tell this from the curves.

I actually imported and hooked up that exact curve (here it is again [55K]) to the geometry. Here's the result. (112K) Well, her mouth moves! Unfortunately, the curve doesn’t quite match the geometry. It was good enough for me to continue onward though. (That gold stuff painted on her head is part of the hair system which is turned off since it takes a hundred years to render anything with it on.)

I wasn't satisfied. For one thing, my rotoscoped curve has about 16 keyframes in it. Since it's only a 30 frame animation, that's one keyframe for every two frames!

One of the interesting things that seems to happen in CGI animation is that the number of keyframes converges on the number of frames pretty quickly as you try to get the quality up.

One of the big lessons at ILM on the "Casper" movie was that in order to get the quality they wanted they were setting a keyframe on every frame. That wasn't the idea. The idea is to set a few keyframes and let the computer interpolate.

(Stay with me - I'm leading up to something here.)

Motion capture sets a keyframe at every frame.

Disney-style animators set keyframes wherever they want, and "inbetweeners" (animators in training) fill in the gaps. But that's essentially a keyframe at every frame, because a person is filling in the gaps.

My animation had half of the frames set as keyframes.

Well, I was pretty interested in getting only about four keyframes in my animation. It's one of my philosophies of development that if the computer can do the job, it should. (Interestingly it's also one of my philosophies that before you build a computerized tool to do the job, do it by hand a couple of times to make sure you understand the process.)

So, for this next animation, I used the rotoscoped NURBS animation as a reference, but then hand animated a single oval NURBS curve. I managed four keyframes with my first attempt (not shown) but finally settled on six.

Here it is: (216K) ... the result of several weeks of work and learning and thinking.

(Here's the same thing in wireframe back when it had only four keyframes (414K)- frequently it's easier to see what's going on in a wireframe view.)

Now we're talking! The solid one actually has seven keyframes, because in order to make the entire sequence more natural, I added 10 frames onto the end, where her mouth resets to a neutral position, and that took an extra keyframe. But the "Splashed Another One!" part is just six keyframes. It's also pretty good considering that there is no secondary animation in the eyes or elsewhere.

Yahoo!

You know, everyone, and I mean everyone, who is any good, uses reference material to animate. This naturally leads to thinking about rotoscoping (2D) or motion-capture (3D). Why not just automate the process of converting your reference material into the animation?

The answer, which apparently must be constantly relearned, is that rotoscoped or motion-captured animation always looks wrong. Something subtle gets lost in the translation.

You can take a video reference and show it to someone and of course it looks great. You can rotoscope over the exact same thing and something looks wrong.

Motion-capture can work - the Protozoa people that made Floops proved it over and over. But they built the exaggeration into the performance. And they watched themselves perform the animation in real-time with a headset. And they hired natural performers.

By contrast, on Titanic (the James Cameron movie), House of Moves in Venice used whoever happened to be walking by as mo-cap subjects. A lot of that animation looks like the digital character is walking with a broom stuck up his ass. It's very stiff. (Don't get me wrong - they did some good stuff - like capturing the movement of the clothing - and most people don't even know those are digital characters on the ship. But that's because they're not close-ups.)

So what's the moral of the story? The moral of the story is this - get detailed reference material, study it, copy it even, and then throw it away and do it from scratch.

That gets the best results. IMHO.