How Digital Performances Are Built: Rigs, Physics and Acting

By vaayaa Editorial Team · Subject panel: Isabella Silva (MFA, Design), Roberto Hernández (MFA, Graphic Design)

A believable digital character comes from two disciplines working in sequence: an animator who authored a performance, and a solver that computes the physical consequences. This explainer walks through skeletal rigs and skinning, keyframe blocking, forward and inverse kinematics, motion capture cleanup, position based dynamics for cloth and hair, crowd steering, and the newer data driven layer. It also covers how production divides active character control from passive simulation, how shots move through review, and which skills keep a reel strong.

A character that feels alive is rarely the work of a single technique. It is a rig that responds predictably, an animator who understands weight and intention, and a simulation layer that absorbs everything the animator should not have to hand keyframe. Understanding where those pieces join is the difference between a puppet that moves and a performance an audience believes.

Authored performance versus computed consequence

Animation is authored; simulation is computed. An animator poses a rig, sets keys, and controls timing frame by frame. A solver takes physical rules, a collision environment and forces, then calculates motion nobody keyframed. Nearly every modern production runs both because they answer different questions.

Performance is the animator's domain. Vancouver Film School describes the craft as breathing life into fully rigged digital models, turning geometry into compelling performances, with acting, anatomy and psychology studied to communicate emotion through posture, timing and expression ^[7]. CG Spectrum frames the same role as a 3D animator specialization that produces believable, relatable movement by instilling humanity through expression, pose and body language ^[12].

Consequence belongs to simulation. A cape, a ponytail and a dangling strap react to motion rather than driving it. Houdini's Vellum solver applies a Position Based Dynamics approach across cloth, hair, grains, fluids and soft body objects ^[1]. One framework covers a wide spread of secondary effects that would be tedious to key by hand.

The split is documented in research too. A survey of interactive character animation notes that physics simulation is widely adopted for fluids, cloth and rag doll characters, while commercial applications still rely on kinematics based approaches for actively controlled characters ^[5]. That gap is the practical heart of the field.

Training programs reflect both sides. The Italian video game academy's two-year track prepares students for roles spanning character animator, creature animator, rigging artist, facial animator and technical animator, using Maya, Unity and Unreal Engine across games, cinema and visual effects ^[8]. Entertainment media programs likewise describe tracks designed to emulate a professional studio, moving students from story creation through film completion as part of a team ^[9].

Bones, topology and the limits of a digital puppet

A rig is a control structure layered over a mesh. It generally holds a bone hierarchy, controllers that drive those bones, and a skinning step binding surface geometry so the model deforms. A study built on skeletal algorithms describes the workflow in concrete terms: creating meshes, adding bones, skinning, editing envelopes, then retargeting captured motion onto the character.

Skinning quality caps how believable deformation can look. Weak weight painting produces collapsing elbows, twisting forearms and shoulders that shear under rotation. The U-tad rigging curriculum lists adapting model topology to rigging and animation requirements, building complex rigs, and achieving required realism in deformations as core competencies ^[11]. Topology is not a modeling detail; it constrains what the rig can achieve.

Inverse kinematics is the workhorse for limb placement. The animator moves a hand or foot target and the solver computes joint rotations that satisfy it. Forward kinematics chains rotations parent to child instead. Hong Kong's data driven computer animation course lists forward kinematics, inverse kinematics and skinning among foundational topics alongside keyframe animation and motion capture ^[2].

Facial setups are their own discipline with their own vocabulary. Blendshapes, musculoskeletal models, action units and face rigs all appear as named approaches in that same syllabus ^[2]. A production character may combine a joint based jaw, corrective shapes at the lip corners and wrinkle maps driven by expression intensity. Every layer is authored, not solved.

Interdisciplinary programs treat that whole stack as one arc. One simulation, animation and gaming curriculum combines traditional art media, programming, creative writing and story development with coursework in modeling, rigging, animation, lighting and camera work in three dimensions ^[10]. The point is that a rigger, an animator and a lighter share vocabulary before they divide labor.

Blocking poses and the acting decisions behind timing

Keyframes remain the backbone of authored performance. An animator sets extreme poses, adds breakdowns, and refines timing on curves. The craft lives in choices: where weight shifts, how long a hold lasts, whether anticipation reads before the action. Keyframe animation leads nearly every syllabus, appearing first in the basics list of the Hong Kong calendar entry ^[2] and sequenced beside motion capture in graduate study ^[6].

Blocking proves a shot before polish. Animators stage key poses that sell the performance, then animate gestures, expressions and timing to match dialogue and emotional beats, as VFS describes the typical responsibility list ^[7]. A blocking pass stays deliberately rough, stepped or lightly splined, chosen so silhouette and read can be judged.

Iteration is where shots improve. VFS notes that animators revise short shots many times, working with directors and supervisors to reach authenticity ^[7]. CG Spectrum places a matching gate in the pipeline: trajectories are rendered as a playblast, which supervisors, directors and clients review for approval ^[12]. Small timing shifts change meaning, which is why gates exist.

Reference is a craft tool, not a shortcut. Character animators act out scenes to camera or rely on captured motion to replicate human movement ^[12]. The camera reference and the final animated performance remain separate deliverables, because a shot's staging may demand a different pose than the one a performer found.

Particles, springs and solvers for passive motion

Physics based methods compute motion from forces. Rigid body dynamics, mass spring systems and particle systems appear together as the standard physics toolkit in graduate animation study ^[6]. A mass spring network models cloth or flesh as point masses joined by springs; particle systems handle dust, spray and large scale effects.

Position based dynamics now dominates production cloth and hair. Instead of integrating forces step by step, the solver projects positions until constraints are satisfied, which is easier to stabilize and to direct. Vellum's documentation states the method is used for cloth, hair, grains, fluids and soft body objects ^[1]. The same family can swing a cape, whip a ponytail and collapse a pile of sand.

Simulation covers what animators should not fight. The literature describes physics as offering responsive, realistic animation, with adoption well documented for fluids, cloths and rag doll characters ^[5]. The payoff is credible weight: fabric that hangs correctly, hair that lags a head turn at a steady rate.

Control is the hard part. A solver is unusable if it cannot be directed. The survey frames two decades of progress around controllability, efficiency, flexibility and visual fidelity all improving together ^[5]. In practice artists pin vertices, animate colliders, adjust constraint stiffness and apply per frame forces, then cache the result.

Textbook coverage has expanded to match. The standard animation programming reference notes expanded treatment of fluids, hair and crowd animation, plus extensive new material on clothes and cloth, and describes the mathematical and algorithmic foundations that let an artist control each technique rather than merely trigger it ^[3].

Capture, retargeting and the cleanup that follows

Motion capture delivers realistic base timing quickly. The skeletal workflow captures real motion data, optimizes it, then retargets it onto a character. Retargeting is unavoidable because capture volumes and character proportions never align exactly.

Raw capture is a starting point. VFS lists integrating and refining captured data so it feels natural and story driven, alongside blocking and expression work ^[7]. CG Spectrum notes performers are sometimes filmed acting out scenes to supply movement reference ^[12]. Reference and delivered performance are separate deliverables.

Cleanup is artistic judgment. Sliding feet, over-rotating wrists and floaty contacts are typical artifacts. Animators fix contacts first because they ground everything else, then restage the upper body for the chosen camera. Timing may be retimed substantially when the edit demands a different beat.

Facial capture follows the same logic at higher stakes. Course content pairs video based capture with blendshapes and action units ^[2]. Higher fidelity raises the bar for cleanup, because audiences notice micro-expression errors on faces more sharply than body motion errors.

Crowds, grooming and scaling detail across a shot

Crowd simulation is its own algorithmic problem. Research equips agents with synthetic vision so steering reflects how people perceive and respond to surroundings; one approach replaces purely reactive collision avoidance with exploring the full range of adaptations and keeping the locally optimal one via a perceptual cost function ^[4]. The result reads less mechanical than simple repulsion models.

Crowd work appears in academic and tooling contexts alike. Simulation of natural phenomena including crowd simulation and deformation of pliant materials is listed in graduate calendar content ^[6], while Houdini's documentation dedicates a section to creating and simulating crowds of characters ^[1].

Grooming and soft tissue scale the same way. Hair and fur, feathers, and muscle, tissue and skin simulation each get dedicated treatment in production documentation ^[1]. Each system adds cost, so a hero creature's feather groom may demand compute that a distant crowd at reduced scale cannot justify.

Budgeting is therefore a creative decision. Supervisors choose which characters carry simulation detail and which stay simple, so shot breakdowns list per asset simulation tasks instead of applying every system to every model.

Where learned control fits and where it stops

Learning based methods now sit inside the standard course map. Graduate study names learning based animation methods, neural network approaches and diffusion models among current topics ^[6]. The Hong Kong outcomes require students to understand and implement learning from motion data or physical animation data ^[2].

Data driven control shifts workflow rather than replacing it. A controller trained on motion data can produce plausible locomotion given a path and speed, instead of hand placing every footstep. Data driven character control, cloth animation, facial animation and skinning all appear as named syllabus topics ^[2].

Generalization is the caution. A model trained on one movement style may not transfer cleanly to a stylized character with different proportions. Teams keep hand authored controls as fallback and deploy learned systems where coverage is proven.

The practical test is always the same. Can the shot be adjusted on request without redoing the whole solve or retraining the whole model? If not, the new method stays a research tool rather than a production one.

Frequently asked questions

Conclusion

Strong digital acting comes from a clear division of labor: an author with rehearsed intent animates the primary action, and a solver computes the consequences. Mastering both sides of that line, instead of drifting toward one, is what lets a character read as alive from a thumbnail check to a final render.

References

1. SideFX. Houdini 22.0.
2. Department of Computer Science, HKU. Courses Offered.
3. searchworks-lb.stanford.edu. articles+ search results - Stanford University.
4. librarysearch.temple.edu. Library Search - Temple University.
5. CNRS. Interactive character animation using simulated physics.
6. Simon Fraser University. Course Outlines - Fall 2025 - CMPT 766 G100.
7. Vancouver Film School. Character Animator.
8. AIV. Corso di CHARACTER ANIMATION.
9. IDP Hotcourses Abroad. Emerging Media (BFA) – Character Animation Track.
10. catalog.emich.edu. Program: Simulation, Animation and Gaming [BS] - Eastern Michigan University.
11. U-tad. M.A. in Rigging & Character FX.
12. CG Spectrum. Character Animator Job Description: Salary, Skills & Career Paths.
13. MAAC Jayanagar. AVGC Salary Guide 2026: VFX & Gaming.