Face Rigging through Curvenet Parametrization

How to read this

Category

production talk, facial rigging parametrization technique

Contributions

• Replaces NURBS patch face parametrization with curvenets, free form curve networks, as the sparse authoring proxy, removing the rectangular grid layout restriction of NURBS patches
• Introduces a surface based interpolation solver that propagates sparse curvenet point weights into weight maps over the full face mesh by minimizing a Laplacian regularized objective, factored once and reused across multiple weight maps
• Adds parametrization coordinates beyond standard uv, including a radial coordinate from the lips outward and a depth coordinate toward the throat, plus an auto smooth tagging system so a rigger only has to hand weight a few points
• Deployed in production on Pixar's shared human face rig and on animal characters in Hoppers (2026), where a new character is rigged by refitting predefined curvenet points rather than rebuilding the parametrization from scratch

Context

This talk extends the curvenet primitive introduced in Character Articulation through Profile Curves (de Goes et al, 2022), which used curvenets as a Presto rigging primitive driving deformation through the Profile Mover deformer, and it sits alongside Shaping the Elements: Curvenet Animation Controls in Pixar's Elemental (Nguyen et al, 2023), which used curvenets for shape controls. Where those works pose geometry directly with curvenets, this talk repurposes the same surface aware interpolation machinery to solve a different problem, generating the face parametrization weights themselves, extending curvenets from a shape control tool into a face rig authoring tool.Builds on: Shaping the Elements: Curvenet Animation Controls in Pixar's Elemental · Character Articulation through Profile Curves · Fine Wrinkling on Coarsely Meshed Thin Shells

Correctness

This is a three page SIGGRAPH Talk, not a full paper, so there is no formal evaluation, comparison table, or user study. Its credibility rests on production deployment: it shipped on multiple animal characters in Hoppers and is now part of Pixar's shared human rig pipeline. The interpolation quality claims are shown only through figures and a supplemental video, not quantitative metrics.

Clarity

Short and readable in one sitting for anyone with prior context on de Goes's Profile Curves work and the Presto rigging pipeline, but it assumes familiarity with NURBS patch parametrization and curvenet primitives, so it reads best for riggers already inside a Pixar style rigging system rather than outside readers.

How to read it

First pass, five minutes: read the abstract and look at Figure 1 for the character faces and Figure 2 for the curvenet weights versus surface weights, enough to get that curvenets replace NURBS as the sparse proxy for face weight authoring. Second pass: read Section 2 in full for the weight object primitive, the S and B matrices, and the Laplacian regularized objective that turns sparse curvenet weights into per vertex weights. A third pass is worth it mainly for riggers actively building a curvenet based pipeline of their own, since at that point reading the cited de Goes 2022 paper in full is likely more valuable than a third pass on this short talk.