Fluid's Michael King, with contributions from Seth Piezas and Michael Fu, shares details on CGI pipeline organization.
Whats the Big Deal?
So what's so tough about making a fully animated CG film? At worst, it's three or four half-hour episodes back to back, a friend once said to us. Now the friend is making a fully animated CG film and ripping out his hair.
Whats so tough? Size matters.
A full feature computer animated film is a beast of enormous scale in all respects. For instance, some typical numbers for todays CG films involve the creation of over 1,000 models, 2,000 shaders, 4,000 texture maps and 100 characters. A crew of around 150 artists and technicians work on assets that are shared over an average of 1,500 shots (Toy Story had 1561) all in various stages of completion. Complex technologies must be developed, released, and maintained while the film is in active production. A creative team can and will request big changes late in the game despite the best laid plans. Stretch the production over two and a half years and you begin to see the scope of the challenges ahead.
Many articles and books have already covered the tasks and departmental structure involved in the creation of CG imagery, but few have illuminated the pros and cons of how these systems actually perform in practice. This article focuses primarily on the organizational principles to keep in mind when producing a CG feature film such as Dinosaur, Shrek or Finding Nemo.
The Task Ahead
We should review what were trying accomplish before we can talk sensibly about how to organize it. A film is built out of shots, so the CG creation process is ultimately designed to produce finished shots. There are any number of ways to craft a shot, but most involve the same fundamental tasks. The tasks can be performed in chronological order, so each task builds upon the result of the previous one.
Heres a simplified example to illustrate the work ahead. Pretend our crew is just one person and our movie consists of one shot of a robot juggling chainsaws in an empty room. What are the steps involved with making this shot in CG?
First, we design and explore the look of our robot, room and chainsaws by producing concept art. Then we provide construction details via model sheets. From these plans, we model moving the 2D drawings by sculpting 3D geometry on the computer. Once the geometry exists, rigging allows the model to move and deform. In this case, our robots arms must be rigged to perform the juggling, and the chainsaws chains should rotate convincingly. Shading applies surface color and textures to the models and adjusts how surfaces respond to light. We want our robot to be shiny metal, the room to be wood paneled and the chainsaws to seem greasy, dirty instruments of destruction.
After our assets are built, were ready to put together the shot. We layout our shot by putting the robot and chainsaws into position in the room, adding the camera, and adjusting the arrangement until the composition works. Animation makes the robot juggle and the chainsaws fly through the air.
Lighting illuminates the room with virtual light and shadow, and adds a soft bloom to the robots highlights. Effects adds exhaust from the chainsaws and sparks when the robot drops a saw on its foot.
Now we can render the shot, which processes all of the data and computes the final images. We perform finishing by retouching small flaws, color correcting and making any additional tweaks on the images.
This description drastically simplifies the CG production process, but its enough context for our discussion about scale and team structure. Start with this small example, but now imagine we must complete 1,500 shots with a crew of 150 to 200 people. Thats just the beginning of the growth in complexity, because were not making 1,500 identical shots. Instead, each shot is unique, though each shot often shares elements with many other shots.
Technology Doesnt Make a Movie. People Do.
Labor is the single most significant expense of below-the-line production costs and is also one of the easiest to misdirect. Once the creative goals of production are established, logistics play a major role in staying within budget and delivering on time.
In the end, moviemaking comes down to unglamorous things like budgets, schedules, communication, and a lot of hard work basic project management. Your artists and technicians must be organized in a way that works for both themselves and the creative team (director, story, etc). These shows usually have a lengthy production cycle with a heavy emphasis on the end. Organizing the crew properly requires an approach that makes efficient use of resources and prevents crew burnout.
As of today, two basic systems have evolved to solve these problems: the assembly line method and the multi-function team method.
The Assembly Line
Think about making an animated feature, and chances are youll think of this approach, which has its roots in the hand-drawn animation industry. In this system, specialists perform one task each, and departments are formed around the tasks. A modeler just builds geometry, an animator just animates and so forth. Each shot in the film travels from one department and artist to another until completion.
It makes logical sense. Create departments around the tasks and staff them with people who are expert at one task. The shots will turn out great, because there is a small army of specialists polishing each shot into perfection. Each department performs a distinct task, which simplifies production planning. When the pipeline is full and all departments are busy, the resulting quality and productivity is amazingly high.
Reality, of course, is messier.
Because the shot work is split between many people and departments, production coordinators face many daunting challenges. Shots can be in wildly different stages of completion in different departments, and some tasks may not fit into the departmental structure and get lost. The pipeline can stall individuals or whole departments may sit idle if the previous stage misses a deadline and theres no give in the schedule.
The artists themselves also face some hurdles. Back to our juggling robot. Tasks in CG can be performed independently, but they are still very interrelated. If a modeler builds a great-looking robot, but ignores shading issues, then that model will come back for rebuilding. If a shading artist makes the metal robot super-reflective, that means more work for the lighting artists. If issues like this are forgotten, then a lot of fixes become inevitable as the shot evolves. To prevent potential problems requires not only tight communication, but also that the artists understand the needs of their peers.
Another pitfall is artist myopia. Without proper information, an artists efforts may never be used. What if the robot were only seen from the waist up? Without that information, the modeler may spend weeks detailing the robots feet without ever knowing that the effort is being wasted. Or the shading artist may spend weeks on a detailed texture for the chainsaws, but the chainsaws are always moving, their motion blur turning them into vague smears of color. Ensuring that each artist understands the context of their work adds overhead to production manager and supervisor responsibilities.
Seth Piezas (left) and Michael Fu (right) co-founded Fluid.
The Multi-Function Team
As CG software became oriented towards artists instead of programmers, individuals began to be proficient in many disciplines. Today one artist is capable of finishing modeling, shading, etc. quickly enough to be cost effective. The visual effects industry pioneered this approach with the notion that a small team of artists could be responsible for all aspects of the shots within a given sequence. The team consists primarily of generalists capable of performing all required tasks. In the interest of efficiency, multiple teams operate in parallel on different sequences.
If all tasks are being worked on in an organic manner, then production management is faced with a conundrum: track the tasks or track the shot? Task tracking is difficult, because tasks are less formally structured in this environment. If the same artist is modeling, shading and rigging our robot, progress and status depend greatly upon the artists idiosyncratic techniques. However, monitoring the shot as a whole may not be informative enough about the state of the entire production.
Having a smaller team work on related shots reduces some of the communications overhead. Each artist understands the needs of the shot, since they are more involved from beginning to end, which makes it easier to focus work on the areas yielding the most return. Also, task interdependence becomes less of an issue when a single artist deals with many tasks. If lighting and shading are considered simultaneously, some interesting shortcuts can occur in the shading that would be difficult to coordinate if the shading and lighting artists were separated by time and department boundaries.
Some gains are lost because multiple, independent teams must coordinate. If our robot is shared between two teams, the needs of both sequences must be considered. If the robot is modeled by one team who only needs it from the waist up, will it be usable by the next team who needs to see the whole robot? If an artist on one team comes up with a lighting technique, how will other teams know about it? In the heat of production, its all too easy for the teams to become fragmented and isolated. Efficiently sharing information and assets between teams becomes a bigger challenge than expected.
A Shot Is Still a Shot
To reach the final level of polish takes about the same amount of time and effort, regardless of the method of organization. If it takes eight weeks for one person to finish our juggling robot shot, then chances are it takes the assembly line about eight weeks as well. In the multi-function team approach, a single artist can work on this shot alone and so calendar time equals actual effort. In the assembly line approach, the artists involved will spend less than eight weeks of effort working on this shot to reach the same result. But the calendar time required is still about eight weeks or longer due to approvals, coordination and hand-offs between departments.
The level of completeness at any point in time is also impacted by the organization. In the multi-function approach, it is much easer to rough out an entire shot with unfinished models, temporary lighting, and such. This allows the shot to be evaluated in a broader context, the creative team makes more informed decisions, and the artist iterates with more knowledge about the important elements to touch. The shot work is front-loaded: broad strokes can be completed in the first 50% of the schedule.
In the assembly line process, each task can go through a rough pass for comments, but context about the whole shot is often missing. The creative team will not have as much context as theyd like until later in the production cycle, typically somewhere between animation and lighting. This leads to individual elements in a shot being more polished than they might need to be, due to artists preparing for worst-case scenarios. The shot work here is back-loaded: there are few broad strokes (except perhaps for previs) and the shots full context is not apparent until the last 50% of the schedule.
Real World Strategy
Now lets apply these basic options to some examples taken from actual projects. This is not to say how these films were actually produced. Were just using examples that will be familiar to most people and illustrate how the different approaches work in various situations.
Pretend we are asked to produce a sequence similar to the sequence from Shrek where the fairy tale folk invade Shreks home in the swamp. This sequence happens mostly in and around Shreks home, which we also know will be visited again in a few other sequences. Plus there are many miscellaneous characters required to populate the shots, which will also pop up again in the story.
These are good criteria for the assembly line approach. The models for Shreks home and surroundings can be built once at a fairly high level of detail to allow for recycling in other sequences. Once the house, furniture, trees and other models are finished, we free our modelers, shaders and painters for other work because weve minimized their need to revisit this location. By building and texturing the models for close scrutiny, we incur some overhead by carrying around some extra detail. But this is offset by the flexibility it allows layout, lighting, and other downstream departments, especially since we are going to re-use these elements in other parts of the film.
A subset of our assembly line can be devoted to creating the fairytale creatures. The principle is basically the same as for our set and prop models for Shreks house and the swamp environment. By focusing a small assembly line on building, rigging and shading these creatures, we create inventory that can be used across our film.
A similar approach might be taken for Finding Nemos reef sequences or for the scare factory sequences in Monsters, Inc. These areas are visited multiple times over the course of the story, and so we build them for flexibility and re-usability.
The assembly line is typically the best approach in these situations because of its efficiency at creating a large number of detailed objects in a short period of time. Remember that these elements will be shared among many artists on many sequences. The assembly line approach has an advantage since it can better absorb the overhead required to establish standards, quality assurance, and other factors that dont directly impact the final image, but do have significant impact on the re-usability of an element.
By contrast, the multi-function team is not designed to build every piece of the set in detail. Instead, this teams forte is building a set with just enough detail to fool the camera for a few shots. If instead there was only one shot of Shreks swamp to deal with, the multi-function approach might build very simple tree models that only work from a certain camera angle, or they might just project tree textures onto cards and spend time finessing more important elements in the shot.
This leads us to the types of sequences that would benefit from the multi-function team approach. One unique shot or a series of dissimilar shots are usually the best sorts of projects handled by the multi-function team. Establishing shots, such as the first shot of Sydney Harbor in Finding Nemo, or a montage like the opening sequence to Dinosaur, are good examples of these sorts of sequences.
The opening to Dinosaur has all the right factors to benefit from a multi-function approach. During the course of the sequence we visit a number of unique environments, and each environment rarely lasts more than a shot or two. There is ample opportunity for a generalist to use a variety of techniques such as projected textures, compositing and matte painting to generate the illusion of these complex spaces without actually building them in detail.
Creation of the dinosaurs in these shots is likely better handled by an assembly line approach to create the individual models and animation cycles, especially since we expect to re-use them elsewhere in the film. But the multi-function team may choose to simulate the herds in their specific shots by rendering dinosaurs onto cards and only using a few full 3D dinosaurs. Its all about what works for the shot.
Part of the door chase from Monsters, Inc. where Mike, Sully and Randall chase each other through Japan, France and so forth, is another example of a sequence well suited for the multi-function team approach. These shots also featured unique locations that were not featured elsewhere in the film.
Because the elements in this approach tend to hold up only under certain camera angles and screen sizes, the chance of their being shared between other shots or sequences is slim. Less time is spent on standardizing or documenting the elements, allowing for more time to be spent polishing their look. We take calculated risks: we dont expect to change these shots or to revisit these elements in new shots. If the creative team changes a camera angle, we could lose a lot of work. If the creative team adds a shot or sequence in this environment, we may not have many elements to easily recycle for the new shots.
Where Do We Go From Here?
Chances are that both of these approaches will be used over the course of a production. And although weve presented the approaches as distinct for illustration purposes, in reality you are more likely to encounter some hybrid an assembly line with a multi-function FX sub-team or other such variations.
While the principles outlined above are aimed at places where all work is done in-house, they are still applicable to the recent interest in outsourcing CG work to overseas or other remote facilities while retaining local creative control. This attempt to pattern CG production after developments in the hand-drawn animation industry is still evolving, especially since a standard outsourcing model suited specifically to CG has yet to materialize. These bids to reduce labor costs are ambitious for the near term. The more remote and diffuse the production staff, the more difficult it becomes to organize and retain creative consistency. A project at a single studio is already difficult enough to manage.
CG feature animation is still a young industry, and everyone is experimenting to tailor existing production methodology to fit CGs unique aspects. Your production team should carefully consider their unique skills, personalities, and the goals of its project before choosing a methodology. Every situation is different dont expect any silver bullets. Even production veterans dont expect to navigate these waters without encountering an iceberg or two. But their experience tells them how to avoid the worst perils and survive what cannot be avoided.
Michael King is vp of production for Fluid, a production management and technology company specializing in high-end, computer-animated feature film. He recently departed Pixar Animation Studios, where he was the sequence supervisor for the The Incredibles. Since 1995 he has performed a variety of management, technical and artistic roles. Mike's production credits include Mulan, Tarzan, Dinosaur, Toy Story 2 and Monsters, Inc., among others.
Contributing writers Seth Piezas and Michael Fu hailed from Pixar before founding Fluid.