CoChat

RigAnything: Template-Free Autoregressive Rigging for Diverse 3D Assets

Open access
Liu, Isabella, Xu, Zhan, Yifan, Wang, Tan, Hao, Xu, Zexiang, Wang, Xiaolong, Su, Hao, Shi, Zifan · 2025 · ACM Transactions on Graphics

The DOI matches its Crossref record; 43 of the 43 references it cites were found in a registry or an index; 3 of the 11 sources its claims cite were checked for support.

Abstract

We present RigAnything, a novel autoregressive transformer-based model, which makes 3D assets rig-ready by probabilistically generating joints and skeleton topologies and assigning skinning weights in a template-free manner. Unlike most existing auto-rigging methods, which rely on predefined skeleton templates and are limited to specific categories like humanoid, RigAnything approaches the rigging problem in an autoregressive manner, iteratively predicting the next joint based on the global input shape and the previous prediction.

Overview

CoChat Summary

TL;DR

RigAnything is a template-free autoregressive transformer that generates skeleton joints/topologies and skinning weights for diverse 3D assets by serializing skeletons in BFS order and using diffusion for continuous joint positions. It reports state-of-the-art skeleton and skinning metrics and completes rigging in about 2 seconds per shape on an A100, versus about 120 seconds for RigNet.

Key findings

  • On the RigNet + Objaverse dataset, RigAnything improves skeleton prediction over RigNet: IoU 0.768 vs. 0.456, precision 0.789 vs. 0.424, recall 0.766 vs. 0.591, CD-J2J 0.034 vs. 0.048, CD-J2B 0.035 vs. 0.042, and CD-B2B 0.020 vs. 0.030.
  • On the humanoid subset, it reports IoU 0.886 vs. TARig 0.603 and NBS 0.337, with CD-J2J 0.030 vs. 0.100 and 0.124, respectively.
  • Connectivity prediction improves to 0.965 classification accuracy vs. RigNet’s 0.358, with edit distance 2.150 vs. 12.663 and CD-B2B 0.001 vs. 0.017.
  • Skinning prediction achieves precision 0.825 vs. RigNet’s 0.755 and average L1 0.432 vs. 0.485, though recall is 0.798 vs. 0.828. Ablations show removing joint diffusion drops skeleton IoU to 0.308 from 0.765, and removing normals or pose augmentation lowers IoU to 0.559 and 0.741, respectively.

Methods

  • The model serializes a tree skeleton in breadth-first search order as a sequence of 3D joint positions and parent indices, then autoregressively predicts the next joint with a transformer using hybrid attention: shape tokens use full self-attention, while skeleton tokens attend to shape tokens and use causal attention among themselves.
  • Joint positions are sampled with a diffusion module rather than discrete prediction; connectivity and skinning weights are predicted by separate modules, and the whole model is trained end-to-end with combined joint diffusion, connectivity, and skinning losses.
  • Training uses the RigNet dataset (2,354 models) plus a filtered Objaverse subset; the abstract reports 9,686 high-quality rigged shapes, Sec. 4.2 reports 12,040 after filtering, and Table 2 totals 9,388 categorized shapes. Inputs are 1024-point clouds with normals, max 64 joints, and online pose augmentation with random joint perturbations up to 45 degrees.

Limitations

  • The paper states that the method does not allow control over rig level of detail, relies solely on geometry which can be ambiguous, and does not account for motion styles such as material-influenced skinning; dynamic data are scarce for addressing this.
  • Evident from the text: RigNet comparisons are conducted on rest-posed objects because RigNet supports only rest poses, so arbitrary-pose performance is not directly benchmarked against RigNet.
  • Evident from the text: the reported Objaverse filtered counts differ across the abstract, Sec. 4.2, and Table 2, which complicates exact dataset reproducibility.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3731149

Cites

43 of 43 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

1 Valid · 5 Not valid · 4 Not checked · 74 Own results and interpretations

42 more claims

Create a free account to see all 84 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 17
  • The quote only partly states this claim. 1
  • Probably not a substantial claim. 22

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Research works printed but not linked: 5

  1. Mark Chen, Alec Radford, Rewon Child, Jeffrey Wu, Heewoo Jun, David Luan, and Ilya Sutskever. 2020. Generative pretraining from pixels. In International conference on machine learning. PMLR, 1691–1703.
  2. Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models. Advances in neural information processing systems 33 (2020), 6840–6851.
  3. Peizhuo Li, Kfir Aberman, Rana Hanocka, Libin Liu, Olga Sorkine-Hornung, and Baoquan Chen. 2021. Learning skeletal articulations with neural blend shapes. ACM Transactions on Graphics (TOG) (2021).
  4. Jing Ma and Dongliang Zhang. 2023. TARig: Adaptive template-aware neural rigging for humanoid characters. Computers & Graphics (2023).
  5. A Waswani N Shazeer N Parmar J Uszkoreit L Jones A Gomez L Kaiser and I Polosukhin. 2017. Attention is all you need. In Advances in Neural Information Processing Systems.

Cited by

See all 13

Metadata

Authors
Liu, I., Xu, Z., Yifan, W., Tan, H., Xu, Z., Wang, X., Su, H., & Shi, Z.
Affiliations
  • University of California San Diego (US)
  • Adobe Systems (United States) (US)
  • Adobe Research, San Jose, USA
Venue
ACM Transactions on Graphics
Published
2025-07-27
Volume
44
Issue
4
Pages
1-12
ISSN
  • 0730-0301 (ISSN-L)
  • 1557-7368
OpenAlex
W4412673671
Cited by
12
References
18
Citations by year
2026: 10 · 2025: 2
FWCI
5.22
Type
journal article
Language
en
Open access
Locations
  • ACM Transactions on Graphics (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Topics
  • Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › 3D Shape Modeling and Analysis
  • Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › Human Motion and Animation
  • Physical Sciences › Computer Science › Computer Graphics and Computer-Aided Design › Computer Graphics and Visualization Techniques
Keywords
autoregressive transformers, skinning weights, diffusion models

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.