Dataflow: ILM's Framework for Procedural Geometry Generation, Simulation Authoring, Crowds, and More

How to read this

Category

Production system / pipeline architecture paper

Contributions

• A node-based graph-evaluation system (Dataflow) for procedural geometry, particle and volume simulation, field authoring, hair instancing with attribute control, and crowd simulation
• A C++ core exposed through Python interfaces that runs as a plugin inside commercial DCC tools (Zeno, Maya, Katana, Houdini)
• Evidence of long-term, cross-production deployment at ILM over roughly eight years and dozens of films

Context

An in-house procedural graph framework in the lineage of node-based DCC dataflow systems, generalizing the approach to span geometry, simulation, hair, and crowds within one engine.

Correctness

This is a studio engineering report, not a peer-reviewed method paper; its value is demonstrated by broad production use across many films rather than by controlled comparison, so read it for architectural decisions rather than benchmarks.

Clarity

Accessible at the systems level; a single first pass conveys the architecture and scope without heavy math.

How to read it

Skim once for the system diagram and the breakdown of what each domain (geometry, sim, hair, crowds) plugs into; a second pass only pays off if you are designing a similar plugin-based procedural framework.