A common theme at this years SIGGRAPH was how different studios are
handling the unique demands of shows with ever increasing scope and
complexity. Two examples of this were Method Studios' work on Wrath
of the Titans, and CineSite's work on John Carter. These two
productions had enormous environments and sets that quickly defied
their standard workflow techniques of wrangling data. In both cases
their teams adopted scalable data driven descriptions of the
environments as separately addressable, hierarchical elements managed
outside of a traditional Maya or Houdini workflow. While these
techniques have been in heavy use in CG Feature animation for quite a
while, especially in pioneering work by PDI for wrangling the jungle
environments in Madagascar, the increasing complexity of live action
environments is making these issues imperative to VFX workflow as
well. And both CineSite and Method rose to the challenge with some
inspired answers to some hugely vexing problems.
In Wrath of the Titans, the Kronos sequence involved massive
environmental destruction with dizzying camera flybys. The Kronos
creature literally breaks from the mountainous cliffs that it's made
from. These shots were as thrilling to audiences as I'm sure they
were terrifying to the VFX artists. But Method studios rose to the
challenge by creating in ingenious system of data-driven, resolution
independent scene elements that could be accessed differently to
achieve maximum rendering and animation efficiency, but within a
unified texturing and rendering paradigm that ensured a consistent
look a variable levels of detail.
They broke down the overall model of the huge mountainous environment
into a collection of useable rocky crag shapes called greebles. These
were located both on and within the volume of the mountain and could
be called up when needed, and in the form most appropriate to that
particular use case. For instance greebles close to camera would be
called up at their highest resolution and those farthest from camera
at there lowest. Taking this hierarchical methodology one step
further, even the rock face texturing was handled as volumes which
could blend regions of differing resolution together. Because the
data locations for each greeble could be distributed within the volume
of the mountain, the actual geometric assets did not have to exist in
the scene until the overlying rock faces crumbled away to reveal them.
Likewise on John Carter of Mars huge data savings were had based on
instanced reuse, but in this case, the instancing itself was leveraged
to provide the actual animation. The large machine environment of
Zodanga was a walking city on centipede like legs. To illustrate the
ingenuity with which CineSite addressed problems of scale we can look
at the legs of the city. Each leg is essentially a piston and the
city walks by articulating these in sequence like a centipede. One
library animation of a single piston cycle was stored as a cache and
then new instances were created from this cached animation database
offset to their locations under the city as well as offset in time.
This meant that new firing sequences could be choreographed without
reanimating hundreds of individual animations, likewise changes to the
animation of the underlying cached cycle would be automatically
inherited by the instances. It also meant that the actual geometry of
the piston animation only had to be stored once for the single
canonical piston animation. Cine Site wrapped all of this
functionality in a very nice, user accessible interface both in
standalone and Maya hosted forms; this allowed artists to access sub
components of the structures for specific effects on a granular level
during shot production.
Because the CineSite team stored all of this data existed in a
queryable object database, it could be programmatically filtered to
check for visibility, set render quality levels and perform other
manipulations that helped to achieve maximum efficiency in terms of
rendering, disk storage and artist interaction speed. Not being
satisfied to sit on their laurels, CineSite is already looking into
making the system more robust by evaluating new technologies like
Alembic, and Renderman's new instancing capabilities to make their
approach even better.