Computer-animated movies

Pipeline, techniques, technology

Saty Raghavachary, DreamWorks Feature Animation

Some history

In the "early days" (mid-70s, 80s), computer animation was confined to pioneering schools such as Utah, Ohio State, NYIT, etc. and cutting-edge studios such as ILM, Pixar, MetroLight Studios, Rhythm & Hues, PDI, RezN8.

During the early and mid-90s, with movies such as Terminator 2, Jurassic Park and Toy Story, "CGI" visual effects and computer animation went MAINSTREAM.

These days, CG animation is "everywhere" - movies, games, commercials, web, ride films, rendered on a variety of hardware - traditional PCs, game machines, cell phones..

Production pipeline - details

A typical computer-animated movie (eg. Shrek 2, Up) requires:
  • about 400 people
  • about 4 years
  • hundreds of thousands of 'assets' [little pieces of the movie]
  • tens of thousands of hours of computer calculations
  • ...

Careful planning is needed to pull this off. A "production pipeline" is ESSENTIAL to keep track of things and make everthing flow smoothly.

An animated movie's production pipeline comprises:

  • stages (departments)
  • tools, processes and workflows, with well-defined inputs and outputs
  • departmental specialists
  • overhead personnel

Here is a detailed rundown of the various departments (with very different skill sets, educational backgrounds, workflow etc.) that make up our beloved DW production assembly. It shows that animated moviemaking (like its live-action counterpart) is incredibly collaborative.

Pipeline - interconnections

As you can see, there are specific pathways for assets (completed animation, laid out scenes, surfaced geometry etc.) to be passed from stage to stage (dept. to dept.).

A pipeline-based workflow allows us to work on multiple productions at once, via staggering and careful timing - our production schedules look EXTREMELY detailed and span large periods of time.

Summary: story in, movie out!

"How we do it" - DW/Pixar/Blue Sky

Pipelines for animated movie production are fundamentally similar across CG studios!
DreamWorks [click on Studio -> Animation 101, use arrow keys to navigate]


Pixar


Blue Sky

*Computer* animation

We're specifically talking about computer animation.. What algorithms, computational techniques etc. go into the creation of the imagery?

Turns out that from modeling and layout to animation and effects to rendering and compositing, algorithms and computation are everywhere.

Transformations

A variety of transformations - deformation by spline, texture mapping, reflection mapping, anisotropic shading..

Subdivision

Geometry is subdivided, to create smoothness, fracture..

In recent shows we've used a 'volumetric' fracture technique that is able to create realistic-looking, hollow and closed fractured shapes out of regular polymeshes.

Tai Lung escapes [1:50]

Fractals

Recursive branching of a tube structure produces corals.

Fields

The classic CG 'blobby' algorithm deals with spatial summation of electron fields. We also use fields to affect particles in particle systems (eg. add a turbulence field, wind field etc.).

Another example of blobby (also called isosurfaces or metaballs) fluids.

Rigid body dynamics

Collisions involving multiple objects cannot be hand animated! Their multi-body interactions need to be simulated, ie. painstakingly computed.

Filters

Multi-pixel (and supersampled) filtering is used to ensure smooth, anti-aliased renders.

Randomness

Random numbers are widely used in animation, to impart a 'natural' look to textures, crowds, motion, etc.

Pseudo-randomness

LDS are better-suited for QMC algorithms compared to PRNs [a TLA soup!]. Here is an example. Also, is 'pseudo-randomness' a pseudo-oxymoron?

Interpolation

A wide variety of quantities can be interpolated in a variety of ways.. Interpolation is another computer animation staple.

PDEs

FX animation almost always involves simulation, which in turn involves 'engineering' type of mathematics (calculus of continuous variables). These classes of problems often involve large grids of data, which can be processed in parallel, SIMD-style. That makes them suitable candidates for GPU acceleration.

Cloth is usually modeled as a series of interconnected springs, on which forces act. Solving for vertex positions at discrete time steps is what the simulation is about, this produces realistic results.

For the 'ultimate' in fluids, check out work by Ron Fedkiw at Stanford.

Getting into CG

Can't choose between art & technology?
Do both - become a TD (Technical Director)!

Resources

Get REALLY good at computer graphics, math and programming!
Graphics


Math [including 'recreational mathematics']


Maya, MEL

Processing

Python, C++, OpenGL, RenderMan Shading Language..

Download FREE software such as Blender and Aqsis.

Enroll in courses from schools such as Gnomon, TD College, Animation Mentor.

Fun