Secondary motion — the jiggle after a stop, the wave that ripples down a tail, the little overshoot in a cartoon ear — is usually one of two things in character animation: hand-keyed by an artist, or produced by an actual physics simulation. Dynamic Skinning, published in IEEE Transactions on Visualization and Computer Graphics (TVCG) 2026, proposes a third option that borrows the appearance of physics without the cost of computing it.
How it avoids the simulation
Traditional rig skinning methods like Linear Blend Skinning (LBS) compute where each vertex sits based purely on the rig’s current pose. Dynamic Skinning adds a layer of deformers that offset each vertex’s position as a function of that vertex’s position derivatives and time — essentially reading how a point is already moving and extrapolating a plausible secondary motion from it, rather than solving the differential equations an actual cloth, soft-body, or rigid-body simulator would use. The result produces convincing oscillation and wave-propagation effects, among others, computed directly rather than iteratively solved — which is also why it’s “extremely fast,” in the authors’ words, compared with simulation-based approaches doing comparable work.
Because the deformers are a direct, closed-form computation rather than a simulation being tuned toward a target look, the technique keeps a high degree of artist control: parameters can be pushed toward a specific cartoon exaggeration rather than toward physical accuracy, which is usually what a character animator actually wants. The paper positions this as a deliberate design goal — supporting the kind of hands-on, iterative workflow a rigger already uses, rather than asking them to tune simulation parameters indirectly and wait for a solve.
Why “no simulation” is the whole point
Physics simulation for secondary motion is powerful but expensive and often fights the animator rather than helping them — a soft-body solve doesn’t know that a director wants the ear to overshoot more for comic effect regardless of what mass and stiffness would actually produce. By computing offsets directly from motion the rig is already producing, Dynamic Skinning sits in a practical middle ground: faster than simulation, more automatic than hand-keying every secondary beat, and more controllable than either.
The paper, accepted to TVCG 2026, is the authors’ accepted version rather than a conference-only preprint — a peer-reviewed publication rather than an early, unreviewed arXiv posting.
Related Reading
- Dynamic Skinning: Kinematics-Driven Cartoon Effects for Articulated Characters — arXiv
- RLMuscle trains a 194-muscle character to move using reinforcement learning
- Getting started with Blender rigging: armatures from scratch
- Getting started with Mixamo: auto-rig and animate a character without hand-weighting a bone