DCC production guide

Skinning and Deformation Methods Guide

Direct answer: Use linear blend skinning as the predictable default, then introduce dual quaternion skinning or corrective shapes only where volume loss is visible and the runtime supports them. Keep influence limits, normalization, bind pose, and skin-cluster settings explicit. Constraint weights control rig relationships; they are not a substitute for deformation weights.

Skin clusterLinear blendDual quaternionCorrectivesRetargeting

1. Define the deformation contract

List target platforms, maximum influences per vertex, supported skinning methods, skeleton naming, bind pose, and corrective-shape support. Test with the actual runtime shader rather than assuming the DCC viewport matches it. The armature or skeleton should have stable joint orientation and scale before binding. Retargeting requirements also affect proportions and twist-joint placement, so settle them before detailed weight painting.

2. Configure the skin cluster

Bind only intended deform joints and remove helper constraints that should not influence vertices. Choose normalization behavior, maximum influences, dropoff, and bind method deliberately. Keep weights non-negative and normalized unless the destination explicitly supports another scheme. After adding or removing joints, audit locked influences and the bind matrix. A skin cluster with hidden stale influences can export differently from what the painted viewport suggests.

3. Compare linear blend and dual quaternion

Linear blend skinning averages transformed positions and is widely supported, but shoulders and twisting forearms can lose volume. Dual quaternion skinning preserves rotating volume better, yet can bulge around some bends and may not match every engine. A blended per-vertex mode can target problem regions where supported. Always compare methods on the same pose set; a relaxed bind pose hides the differences that matter.

4. Paint weights with constraints

Start with rigid ownership around stable bone regions, then smooth only across genuine anatomical transitions. Respect the engine influence cap while painting instead of pruning at export. Constraint weight belongs to parent, point, or orient relationships that drive controls or joints; verify that its animation does not create double transforms. Mirror weights only after left-right naming and joint orientation are consistent, then inspect the center seam manually.

5. Add shape-based correction

Use a blend shape or shape key for deformation that joint weighting cannot solve cleanly, such as shoulder compression, elbow volume, or facial wrinkles. Drive correctives from joint angles or pose-space logic and keep their sculpt deltas localized. Confirm evaluation order: a corrective before skinning behaves differently from one after skinning. Export support varies, so bake or reproduce the driver logic when the destination cannot carry it directly.

6. Decision example: twisting forearm

A forearm collapses during a 150-degree twist. First distribute rotation across twist joints and repaint the transition using linear blend skinning. If volume still collapses and the engine supports it, blend dual quaternion influence through the middle of the forearm while retaining linear behavior near the wrist. If the silhouette still needs a precise anatomical bulge, add a small driven corrective instead of broadly inflating weights.

7. Diagnose deformation failures

Candy-wrapper collapse indicates concentrated twist or linear blending across too few joints. A sudden vertex spike usually means an unnormalized, negative, or unintended influence. Mesh movement when a control constraint changes can reveal double transforms. A shape popping at one angle points to a discontinuous driver or wrong evaluation order. Retargeted shoulder distortion often comes from mismatched rest proportions, joint orientation, or missing twist distribution.

8. Evidence checklist and multilingual workflow table

Capture influence-count checks, normalized-weight checks, bind-pose restore, and deformation poses for shoulders, elbows, wrists, hips, knees, jaw, and eyes as applicable. Include runtime comparisons for each supported skinning method, plus screenshots of skin-cluster and corrective settings. Record the exact skeleton version and retarget profile. The multilingual rows below connect rig structure, skinning choices, and shape correction in review notes.

中文English日本語Workflow use
约束权重Constraint WeightコンストレイントウェイトAudit rig-driving relationships separately from vertex weights.
双四元数蒙皮Dual Quaternion SkinningデュアルクォータニオンスキニングUse selectively where rotation must preserve volume.
线性混合蒙皮Linear Blend Skinning線形ブレンドスキニングKeep as the interoperable baseline.
蒙皮簇Skin ClusterスキンクラスタRecord bind matrices, influences, and normalization.
骨架ArmatureアーマチュアConfirm Blender rig object and bind state.
混合形状Blend ShapeブレンドシェイプApply localized pose or facial correction.
重定向RetargetingリターゲットValidate rest proportions and mapping.
骨架SkeletonスケルトンVersion the exported joint hierarchy.
形态键Shape KeyシェイプキーTrack Blender corrective or facial targets.

Related production references

Open SEELE AI workspace

FAQ

Is dual quaternion skinning always better?

No. It preserves twist volume but can bulge and is not supported identically everywhere. Compare it with the target runtime.

Are constraint weights the same as skin weights?

No. Constraint weights blend rig relationships, while skin weights assign vertex influence to deform joints.

When should I use a corrective shape?

Use one after joint placement and weight painting cannot achieve the required silhouette across a specific pose.

Why do weights change after export?

Influence pruning, normalization, bind-matrix differences, or unsupported skinning modes can change the runtime result.