TailorRig: Controllable and Structured Auxiliary Skeleton Generation for Garment-Aware Rigging

SIGGRAPH ASIA 2026

Yajing Chen^*†,1, Xin Chen^*,2, Saidan Lu^*,1, Yeqi Yuan³, Weikai Chen¹, Changyu Hu⁴, Cheng Zeng⁴, Xin Wang¹

¹LIGHTSPEED, Tencent ²Xi'an Jiaotong University ³Peking University ⁴Tsinghua University

^*Equal contribution ^†Corresponding author

Abstract

Humanoid rigging for modern characters with complex garments remains challenging due to the need for dense auxiliary skeletons and precise skinning control. Existing automatic methods primarily recover rigs from geometry, providing limited artist control over auxiliary articulation and often lacking explicit semantic or kinematic-chain structure for dense garment bones, leading to instability and weight bleeding.

We present TailorRig, a framework that reformulates garment-aware humanoid rigging as a structured, controllable full-rig generation problem. TailorRig exposes two explicit axes of artist control within a unified framework: a user-specified body skeleton template that preserves body-rig compatibility, and artist-provided semantic labels that drive an autoregressive generator to synthesize requested category-specific auxiliary bone chains. To robustly assign skinning weights under dense skeletal configurations, we further introduce a hierarchical chain-to-bone formulation: vertices are first associated with a kinematic chain, and bone weights are then refined within that chain, reducing weight bleeding between spatially adjacent but semantically unrelated structures.

We evaluate TailorRig on a diverse set of characters and introduce Human-Mix, a large-scale dataset of over 38,000 humanoid models curated for garment-aware rigging. Experiments show that TailorRig improves skeleton accuracy, deformation stability, controllability, and practical usability over existing methods.

Animation Results

TailorRig takes a character mesh, an artist-specified body skeleton template, and a selected set of auxiliary categories as input, and produces a complete skeleton together with skinning weights. The body template provides the first axis of control. It is canonicalized into a standard pose and encoded while retaining its original topology, after which a topology-aware mesh-bone fusion decoder aligns its joints to the target geometry. Because the connectivity of the template is preserved throughout this process, the resulting body rig remains compatible with the artist's intended skeletal convention and can directly reuse existing animation assets.

Auxiliary category labels provide the second axis of control. Rather than treating a label such as skirt, hair, sleeve, or accessory as a global tag, TailorRig grounds it in the current character through reciprocal attention between mesh features and the aligned body skeleton. The resulting geometry-aware label representation conditions a structured autoregressive decoder, which serializes and predicts each auxiliary skeleton tree as an ordered sequence. This formulation allows artists to request any subset of supported categories while generating only the corresponding auxiliary chains and attaching them consistently to the body rig.

After skeleton generation, TailorRig predicts skinning weights with a hierarchical chain-to-bone formulation. The full skeleton is first partitioned into kinematic chains, and each vertex is softly associated with the most relevant chains using learned feature similarity together with geodesic distance cues. Bone-level weights are then refined within this structural context, with each bone prediction conditioned on the weight of its parent chain. This coarse-to-fine organization narrows competition to structurally related bones and reduces weight bleeding between spatially close but semantically unrelated regions, such as arms, garments, and accessories.

Training and evaluation use Human-Mix, which contains 38,088 rigged humanoid characters annotated with a 21-class taxonomy covering 11 body classes and 10 auxiliary categories. Among them, 7,824 characters contain at least one auxiliary category. These annotations provide the semantic and structural supervision needed for body alignment, auxiliary skeleton generation, and hierarchical skinning across diverse character and garment configurations.

BibTeX

Placeholder citation. The final entry will be updated after publication.

@inproceedings{chen2026tailorrig,
  title     = {TailorRig: Controllable and Structured Auxiliary Skeleton Generation for Garment-Aware Rigging},
  author    = {Chen, Yajing and Chen, Xin and Lu, Saidan and Yuan, Yeqi and Chen, Weikai and Hu, Changyu and Zeng, Cheng and Wang, Xin},
  booktitle = {SIGGRAPH Asia 2026 Conference Papers},
  year      = {2026},
  note      = {To appear}
}