Delta Mush: Smoothing Deformations While Preserving Detail

How to read this

Category

production deformer paper

Contributions

• Introduces Delta Mush, a generic post-process deformer that smooths a deformed mesh's motion while preserving the original surface detail and most of its volume
• Works by smoothing both the rest mesh and the deformed mesh with the same Laplacian operator, caching each vertex's offset from the smoothed rest mesh in a local coordinate frame, then reapplying that cached offset in the smoothed deformed mesh's local frame
• Lets riggers use much simpler skin binds and coarser deformers while still getting production-quality results, cutting a ready-to-animate skeletal rig down to about 30 minutes for a senior character TD and 2 hours for a junior one
• Documents its own origin as a production-driven prototype, built inside Rhythm and Hues' Voodoo pipeline under real schedule pressure across Mr Popper's Penguins, Alvin and the Chipmunks: Chip-Wrecked, and Life of Pi

Context

The paper positions itself against corrective shapes and Pose Space Deformation, Lewis et al. 2000, which are effective but expensive to sculpt and tune by hand for every problem pose. Delta Mush replaces that sculpting cost with a cheap, generic smoothing operator applied as a Voodoo Channel. It has since become a standard node shipped in most major DCCs and engines, so most riggers today meet it as a stock deformer rather than as a paper.

Correctness

The claims rest on production experience, it had been in continuous use on all R&H character rigs since 2010 by the time of publication, rather than on formal error bounds. The authors are explicit that Delta Mush does not guarantee equal volume between rest and deformed states, the goal is stated as simple and practical rather than exact. It is also known in practice to over-smooth on sharply twisting joints unless paired with additional bind-pose corrections, a limitation later variants of the technique addressed.

Clarity

A short, clearly written DigiPro practitioner paper, the core algorithm is four compact equations and is understandable to anyone comfortable with basic linear algebra and local coordinate frames.

How to read it

First pass: abstract and Figure 2 to get the core idea, smooth twice, transfer the delta. Second pass: Section 3, the actual algorithm, which is short enough to read in full. Section 4 on Voodoo Channels and Homes can be skimmed since it is specific to that pipeline. Third pass, worth the time even for a quick reader: Section 6's discussion of how the technique emerged from prototyping rather than top-down algorithm design, useful as a case study in production tool development.