Research & Innovation

A Pixar Crowds Veteran Is Co-Authoring a Fix for Animation's Slowest Bottleneck: Cloth

A new neural clothing method, co-authored by Pixar crowds technical supervisor Paul Kanyuk, predicts real-time garment deformation directly from a character's pose — letting an animator see folds and wrinkles update live instead of waiting on a physics solve.

Cloth simulation has been a persistent bottleneck in character animation for decades: get the physics right and a garment moves convincingly, but the solve is slow and the parameter tuning is tedious enough that artists often pose a character blind, then wait to find out what the clothes actually did. A paper posted November 13, Pose-based Neural Clothing for Animated Characters, attacks that problem directly — and one of its eight co-authors is Paul Kanyuk, Pixar’s longtime crowds technical supervisor, whose studio credits run from WALL-E through Elemental.

Learning deformation instead of solving for it

The paper’s framing is explicit about the tradeoff traditional simulation forces: data-driven, learned deformation models offer “a powerful alternative” to physics solves, capable of learning high-quality folds and wrinkles from a dataset of previously simulated cloth. Because the resulting network is lightweight, it can run at real-time inference speed — meaning an artist can interactively pose a character and see plausible, high-quality fold and wrinkle behavior update immediately, rather than kicking off a simulation and waiting.

Why “interactive” is the actual feature

The paper’s own framing is specific about why that speed matters: it’s not just a performance number, it’s a change in how creative decisions get made. An animator who can see cloth deformation update live, in the moment they’re posing a character, can make creative choices about silhouette and motion while posing — rather than discovering, after a slow solve finishes, that a pose doesn’t read the way they expected and needing to iterate blind all over again.

A Pixar name in a research paper is itself a signal

Pixar has a long institutional history of publishing its internal crowd and simulation research at SIGGRAPH — Kanyuk himself has co-authored papers on Pixar’s Presto crowds workflows and scalable volumetric crowds on Elemental. His presence as a co-author on a neural-clothing paper focused explicitly on artist interactivity is a reasonable signal of where that kind of thinking is headed next: not toward replacing an animator’s judgment, but toward collapsing the feedback loop between a creative decision and seeing its actual consequence on screen.