60+
Papers at top venues
30+
Open-source projects
35K+
GitHub stars
Best Paper
at SIGGRAPH
Current projects.
Open source
One Real-Time Generative Renderer, Many Worlds
Read more
World model initiative
Research preview of an interactive world model. Spaces that keep running after you leave them.
Read more
SIGGRAPH 2026
Native Mesh Generation with Diffusion
Read paper
SIGGRAPH 2026
Open-source 3D Gaussian generator. You set the quality budget, and there is almost nothing to install.
GitHub
Open source
Open model, image to mesh. The one most of the ecosystem builds on.
GitHub
Open source · Stability AI
Textured meshes without a GPU. Built with Stability AI; 15,000 stars, developers in 157 countries.
GitHub
Every paper, with its code and weights.
Four papers this year. Weights, code and plugins go out on the day of the paper.
July 19, 2026
Yuqing Zhang, Yan-Pei Cao, Hao Xu, Yiqian Wu, Sirui Lin, Yuqing Wang, Ding Liang, Yuan-Chen Guo, Xiaogang Jin
PixTex performs multi-view diffusion directly in pixel space to generate high-quality, globally consistent textures for 3D assets.
July 15, 2026
Hanxiao Wang, Ying-Tian Liu, Yuan-Chen Guo, Qi-Yuan Feng, Zi-Xin Zou, Ding Liang, Biao Zhang, Yan-Pei Cao
Nexus generates native triangle meshes with diffusion by decoupling vertex generation and topology modeling, avoiding the inefficiency of autoregressive mesh serialization.
June 11, 2026
Qi-Yuan Feng, Zi-Xin Zou, Haoxiang Chen, Yuanchen Guo, Yan-Pei Cao, Tai-Jiang Mu, Shi-Min Hu
Grow3D is an autoregressive 3D shape generation framework that predicts hierarchical octree structures in a coarse-to-fine manner, enabling fast and high-fidelity 3D generation.
June 10, 2026
Cheng-Feng Pu, Jia-Peng Zhang, Meng-Hao Guo, Yan-Pei Cao, Shi-Min Hu
TopoCap extracts motion from monocular videos and retargets it to 3D characters with arbitrary skeletal topologies, enabling topology-agnostic video-to-animation.
June 5, 2026
Qun-Ce Xu, Jiahui Li, Yan-Pei Cao, Weihao Cheng, Tai-Jiang Mu, Ying Shan, Chuan Li, Da Chen, Yong-Liang Yang, Shi-Min Hu
This work addresses fractured object recovery by not only reassembling observed fragments, but also predicting missing parts to reconstruct the complete object shape.
May 8, 2026
Runjie Yan, Yan-Pei Cao, Peng Wang, Ding Liang, Yuan-Chen Guo
TripoSplat introduces Density-Sampled Gaussians (DeG) for fully adaptive, grid-free 3D generation. We achieve differentiable densification by parameterizing primitive centers as samples from a learnable spatial density, optimized directly via a novel render-loss gradient. To model these unstructured sets, our VecSeq diffusion framework resolves permutation ambiguity by anchoring latents to a deterministic 3D Sobol sequence. TripoSplat achieves state-of-the-art single-image-to-3D generation while uniquely enabling variable-resolution decoding from a single compact latent code.
April 9, 2026
Yi-Hua Huang, Zi-Xin Zou, Yuting He, Chirui Chang, Cheng-Feng Pu, Ziyi Yang, Yuan-Chen Guo, Yan-Pei Cao, Xiaojuan Qi
AniGen overcomes the brittleness of sequential "generate-then-rig" pipelines by directly generating fully animatable 3D assets from a single image. We achieve this by unifying shape, skeleton, and skinning weights into a shared continuous spatial representation: S³ Fields. To model rigs of arbitrary complexity, we introduce a joint-count agnostic Dual Skin Field, paired with a confidence-decaying skeleton field that explicitly resolves topological ambiguities at kinematic boundaries. By generating these compressed fields via a two-stage structured latent flow-matching architecture, AniGen guarantees intrinsic structural consistency between geometry and articulation. The result is a robust, end-to-end foundation for generating production-ready, instantly animatable characters and objects across diverse categories.
March 18, 2026
Yang-Tian Sun, Zehuan Huang, Yifan Niu, Lin Ma, Yan-Pei Cao, Yuewen Ma, Xiaojuan Qi
StereoWorld is a camera-guided stereo video generation model that jointly learns appearance and binocular geometry, enabling consistent stereo video synthesis for VR and embodied AI applications.
March 2, 2026
Hanxiao Wang, Yuan-Chen Guo, Ying-Tian Liu, Zi-Xin Zou, Biao Zhang, Weize Quan, Ding Liang, Yan-Pei Cao, Dong-Ming Yan
Autoregressive mesh generation is traditionally bottlenecked by the quadratic compute cost of modeling flattened vertex coordinate sequences. FACE fundamentally resolves this by elevating generation to a higher semantic tier. Through a novel “one-face-one-token” strategy, we encapsulate entire triangle faces into single unified tokens, slashing sequence lengths by a factor of nine. This architectural elegance yields an unprecedented compression ratio of 0.11, doubling the efficiency of prior state-of-the-art without relying on brittle, lossy traversal heuristics. By coupling this highly efficient Autoregressive Autoencoder (ARAE) with latent diffusion, FACE provides a robust, scalable, and compute-efficient foundation for high-fidelity direct mesh generation.
February 4, 2026
Jia-Peng Zhang, Cheng-Feng Pu, Meng-Hao Guo, Yan-Pei Cao, Shi-Min Hu
Current auto-rigging methods fail to scale because they treat skinning as a brittle, high-dimensional regression task decoupled from skeleton generation. SkinTokens solves this by compressing sparse skinning matrices into discrete token sequences. This enables TokenRig, an autoregressive framework that jointly synthesizes skeletal topology and surface deformations as a single coherent sequence. Post-trained via reinforcement learning (GRPO) with explicit geometric rewards, TokenRig generalizes robustly to complex, in-the-wild assets. By seamlessly unifying skeleton and skin prediction, the framework achieves up to a 133% improvement in skinning accuracy, establishing a scalable, end-to-end foundation for animation-ready 3D content.
December 14, 2025
Hao Xu, Yuqing Zhang, Yiqian Wu, Xinyang Zheng, Yutao Liu, Xiangjun Tang, Yunhan Yang, Ding Liang, Yingtian Liu, Yuanchen Guo, Yanpei Cao, Xiaogang Jin
Built on native brick tokenization and backed by our 55,000-model LegoVerse dataset, LegoACE is an autoregressive engine that generates expressive LEGO assemblies from text or normal maps.
December 8, 2025
Cheng Zhang, Boying Li, Meng Wei, Yan-Pei Cao, Camilo Cruz Gambardella, Dinh Phung, Jianfei Cai
UCPE is a unified camera positional encoding that captures pose, intrinsics, and lens distortion in one formulation, achieving state-of-the-art camera controllability in video diffusion while adding under 1% trainable parameters.
December 4, 2025
Chia-Hao Chen, Zi-Xin Zou, Yan-Pei Cao, Ze Yuan, Guan Luo, Xiaojuan Qi, Ding Liang, Song-Hai Zhang, Yuan-Chen Guo
LaFiTe generates seamless textures directly on 3D surfaces through a sparse latent color field, exceeding prior reconstruction quality by over 10 dB PSNR—without any UV mapping.
November 25, 2025
Yangguang Li, Xianglong He, Zi-Xin Zou, Zexiang Liu, Wanli Ouyang, Ding Liang, Yan-Pei Cao
ShapeGen is a scalable 3D shape synthesis framework that improves geometry quality and structural consistency via enhanced 3D representations, higher-resolution supervision, and linear transformer architectures.
August 19, 2025
Feng-Lin Liu, Shi-Yang Li, Yan-Pei Cao, Hongbo Fu, Lin Gao
Sketch3DVE is a sketch-based 3D-aware video editing framework, supporting precise structural manipulation of scenes under large viewpoint changes while preserving multi-view consistency and realistic video generation.
August 19, 2025
Ken Deng, Yunhan Yang, Jingxiang Sun, Xihui Liu, Yebin Liu, Ding Liang, Yan-Pei Cao
Recasting 3D part segmentation as prompt-controllable multi-view 2D mask prediction atop a SAM2 backbone, GeoSAM2 reaches state-of-the-art interactive part segmentation without text input or per-shape training.
July 8, 2025
Yunhan Yang, Yufan Zhou, Yuan-Chen Guo, Zi-Xin Zou, Yukun Huang, Ying-Tian Liu, Hao Xu, Ding Liang, Yan-Pei Cao, Xihui Liu
OmniPart introduces a part-aware 3D generation framework that produces editable and structurally coherent 3D assets through controllable part decomposition and joint part synthesis.
July 5, 2025
Ze Yuan, Xin Yu, Yangtian Sun, Yuan-Chen Guo, Yan-Pei Cao, Ding Liang, Xiaojuan Qi
SeqTex reformulates 3D texture generation as a sequence modeling task, directly synthesizing high-fidelity UV texture maps from pretrained video diffusion priors for consistent and realistic mesh texturing. (arxiv.org)
June 24, 2025
Zehuan Huang, Haoran Feng, Yangtian Sun, Yuanchen Guo, Yanpei Cao, Lu Sheng
AnimaX presents a feed-forward framework for animating articulated 3D assets by leveraging video diffusion motion priors to generate coherent motions across arbitrary skeleton structures.
June 20, 2025
Wang Zhao, Yan-Pei Cao, Jiale Xu, Yuejiang Dong, Ying Shan
Assembler frames 3D part assembly as a diffusion-based generative task, enabling scalable reconstruction of complex objects from input parts and reference images through sparse anchor point representations. (arXiv)
April 16, 2025
Jia-Peng Zhang, Cheng-Feng Pu, Meng-Hao Guo, Yan-Pei Cao, Shi-Min Hu
UniRig develops a unified autoregressive rigging framework that automatically generates high-quality skeletons and skinning weights for diverse 3D assets with complex topologies and categories.
April 10, 2025
Yunhan Yang, Yuan-Chen Guo, Yukun Huang, Zi-Xin Zou, Zhipeng Yu, Yangguang Li, Yan-Pei Cao, Xihui Liu
HoloPart tackles 3D part amodal segmentation through a diffusion-based framework that reconstructs complete semantic parts from incomplete or occluded 3D geometry while preserving global shape consistency.
March 28, 2025
Li-Heng Chen, Zi-Xin Zou, Chang Liu, Tianjiao Jing, Yan-Pei Cao, Shi-Sheng Huang, Hongbo Fu, Hua Huang
GCRayDiffusion enhances pose-free 3D surface reconstruction by integrating geometrically consistent ray diffusion into triplane-based SDF learning, enabling accurate camera pose estimation and robust reconstruction from sparse-view inputs.
March 27, 2025
Xianglong He, Zi-Xin Zou, Chia-Hao Chen, Yuan-Chen Guo, Ding Liang, Chun Yuan, Wanli Ouyang, Yan-Pei Cao, Yangguang Li
Built on the proposed SparseFlex representation, TripoSF supports arbitrary-topology 3D generation and reconstruction by combining sparse voxel efficiency with flexible mesh-based geometry modeling.
February 10, 2025
Yangguang Li, Zi-Xin Zou, Zexiang Liu, Dehu Wang, Yuan Liang, Zhipeng Yu, Xingchao Liu, Yuan-Chen Guo, Ding Liang, Wanli Ouyang, Yan-Pei Cao
Leveraging large-scale rectified flow transformers and a curated 3D dataset pipeline, TripoSG achieves high-fidelity image-to-3D shape synthesis with strong geometric detail and cross-domain generalization.
February 10, 2025
Ying-Tian Liu, Jiajun Li, Yu-Tao Liu, Xin Yu, Yuan-Chen Guo, Yan-Pei Cao, Ding Liang, Ariel Shamir, Song-Hai Zhang
NeuFrameQ proposes a neural frame field representation for anisotropic quadrangulation, enabling scalable and generalizable quad mesh generation across complex 3D geometries with improved directional consistency and topology quality.
December 19, 2024
Wang Zhao, Yan-Pei Cao, Jiale Xu, Yuejiang Dong, Ying Shan
DI-PCG builds a diffusion-driven procedural content generation framework that synthesizes large-scale, controllable 3D scenes by jointly modeling semantic layouts and geometric details for interactive virtual environments.
December 16, 2024
Yi-Hua Huang, Ming-Xian Lin, Yang-Tian Sun, Ziyi Yang, Xiaoyang Lyu, Yan-Pei Cao, Xiaojuan Qi
Deformable Radial Kernel Splatting replaces fixed Gaussian primitives with deformable radial kernels, improving geometric flexibility and rendering fidelity for high-quality neural surface reconstruction and view synthesis.
December 4, 2024
Zehuan Huang, Yuan-Chen Guo, Haoran Wang, Ran Yi, Lizhuang Ma, Yan-Pei Cao, Lu Sheng
By attaching a lightweight adaptation module to pretrained diffusion models, MV-Adapter enables view-consistent multi-image generation for downstream 3D creation tasks.
December 4, 2024
Zehuan Huang, Yuan-Chen Guo, Xingqiao An, Yunhan Yang, Yangguang Li, Zi-Xin Zou, Ding Liang, Xihui Liu, Yan-Pei Cao, Lu Sheng
MIDI-3D formulates multi-instance image-to-3D generation as a diffusion process, enabling coherent reconstruction of complex scenes with multiple interacting objects from a single image input.
November 27, 2024
Jing-Wen Yang, Jia-Mu Sun, Yong-Liang Yang, Jie Yang, Ying Shan, Yan-Pei Cao , Lin Gao
Rather than relying on explicit 3D supervision, WonderWorld generates immersive and explorable 3D scenes from a single image by combining panoramic scene expansion with geometry-aware multi-view synthesis.
November 26, 2024
Yijia Hong, Yuan-Chen Guo, Ran Yi, Yulong Chen, Yan-Pei Cao, Lizhuang Ma
A physics-guided motion generation framework that enhances human-object interaction realism by embedding physical constraints and dynamics into diffusion-based 3D motion synthesis.
November 25, 2024
Ken Deng, Yuan-Chen Guo, Jingxiang Sun, Zi-Xin Zou, Yangguang Li, Xin Cai, Yan-Pei Cao, Yebin Liu, Ding Liang
DetailGen3D enhances coarse 3D generation pipelines with a detail-aware refinement framework, producing high-fidelity geometry and texture details while preserving the structural consistency of the original assets.
November 22, 2024
Xin Yu, Ze Yuan, Yuan-Chen Guo, Ying-Tian Liu, JianHui Liu, Yangguang Li, Yan-Pei Cao, Ding Liang, Xiaojuan Qi
TEXGen adopts a controllable texture generation pipeline that synthesizes high-resolution and style-consistent 3D textures from text prompts while preserving geometric alignment across views.
October 16, 2024
Jingxiang Sun, Cheng Peng, Ruizhi Shao, Yuan-Chen Guo, Xiaochen Zhao, Yangguang Li, Yanpei Cao, Bo Zhang, Yebin Liu
DreamCraft3D++ improves text-to-3D generation by combining hierarchical 3D Gaussian splatting with diffusion priors, enabling detailed geometry synthesis and high-quality texture refinement under complex prompts.
September 16, 2024
Peng Li, Wangguandong Zheng, Yuan Liu, Tao Yu, Yangguang Li, Xingqun Qi, Xiaowei Chi, Siyu Xia, Yan-Pei Cao, Wei Xue, Wenhan Luo, Yike Guo
Personalized human reconstruction from sparse-view images is achieved through PSHuman, which combines diffusion priors with geometry-aware modeling to recover detailed appearance and identity-consistent 3D humans.
August 27, 2024
Bojun Xiong, Si-Tong Wei, Xin-Yang Zheng, Yan-Pei Cao, Zhouhui Lian, Peng-Shuai Wang
Built on octree-based sparse representations, OctFusion accelerates high-resolution 3D shape generation while maintaining geometric fidelity and memory efficiency in diffusion-based modeling.
June 19, 2024
Yang-Tian Sun, Yi-Hua Huang, Lin Ma, Xiaoyang Lyu, Yan-Pei Cao, Xiaojuan Qi
Splatter-A-Video brings 3D Gaussian splatting into video generation, enabling temporally coherent and spatially consistent scene synthesis from dynamic video inputs.
May 19, 2024
Xianglong He, Junyi Chen, Sida Peng, Di Huang, Yangguang Li, Xiaoshui Huang, Chun Yuan, Wanli Ouyang, Tong He
GVGEN explores geometry-aware video generation by jointly modeling scene structure and motion dynamics, producing view-consistent videos with improved spatial coherence across frames.
April 21, 2024
Ying-Tian Liu, Yuan-Chen Guo, Guan Luo, Heyi Sun, Wei Yin, Song-Hai Zhang
A coarse-to-fine 3D generation framework that reconstructs detailed and view-consistent geometry from sparse observations through progressive shape refinement and multi-view diffusion guidance.
March 17, 2024
Tong Wu, Yu-Jie Yuan, Ling-Xiao Zhang, Jie Yang, Yan-Pei Cao, Ling-Qi Yan, Lin Gao
High-quality 3D content generation is achieved through TripoSR, a fast single-image reconstruction approach that combines sparse volumetric representations with scalable neural rendering for efficient shape recovery.
March 4, 2024
Dmitry Tochilkin, David Pankratz, Zexiang Liu, Zixuan Huang, Adam Letts, Yangguang Li, Ding Liang, Christian Laforte, Varun Jampani, Yan-Pei Cao
TripoSR delivers real-time single-image-to-3D reconstruction through a feed-forward framework that recovers detailed geometry and appearance with scalable sparse representations.
February 27, 2024
Hao-Yang Peng, Jia-Peng Zhang, Meng-Hao Guo, Yan-Pei Cao, Shi-Min Hu
CharacterGen presents a controllable 3D character creation pipeline that generates stylized and animation-ready characters from text and image conditions with consistent geometry and appearance.
January 16, 2024
Xin Yu, Yuan-Chen Guo, Yangguang Li, Ding Liang, Song-Hai Zhang, Xiaojuan Qi
This work introduces classifier score distillation for text-to-3D generation, showing that classifier-free guidance can serve as the main optimization signal for high-quality 3D generation.
December 14, 2023
Zi-Xin Zou, Zhipeng Yu, Yuan-Chen Guo, Yangguang Li, Ding Liang, Yan-Pei Cao, Song-Hai Zhang
TriplaneGaussian integrates triplane representations with 3D Gaussian splatting to improve rendering efficiency and geometric quality for large-scale neural scene reconstruction and novel view synthesis.
December 14, 2023
Ying-Tian Liu, Yuan-Chen Guo, Guan Luo, Heyi Sun, Wei Yin, Song-Hai Zhang
Progressive multi-view refinement and diffusion guidance are combined in this work to reconstruct detailed, view-consistent 3D geometry from sparse visual observations.
December 13, 2023
Zexiang Liu, Yangguang Li, Youtian Lin, Xin Yu, Sida Peng, Yan-Pei Cao, Xiaojuan Qi, Xiaoshui Huang, Ding Liang, Wanli Ouyang
A unified motion synthesis framework for generating realistic human interactions by modeling fine-grained body dynamics and temporal coherence under diverse action conditions.
December 10, 2023
Zehuan Huang, Hao Wen, Junting Dong, Yaohui Wang, Yangguang Li, Xinyuan Chen, Yan-Pei Cao, Ding Liang, Yu Qiao, Bo Dai, Lu Sheng
EpiDiff leverages epipolar constraints within diffusion models to produce geometrically consistent multi-view image generation for downstream 3D reconstruction and scene synthesis tasks.
December 5, 2023
Yunhan Yang, Yukun Huang, Xiaoyang Wu, Yuan-Chen Guo, Song-Hai Zhang, Hengshuang Zhao, Tong He, Xihui Liu
DreamComposer enables compositional 3D scene generation by coordinating multiple objects and semantic relationships within a unified diffusion-based creation pipeline.
December 4, 2023
Yi-Hua Huang, Yang-Tian Sun, Ziyi Yang, Xiaoyang Lyu, Yan-Pei Cao, Xiaojuan Qi
Structured and controllable neural scene reconstruction is achieved in SC-GS through semantic-constrained Gaussian splatting, improving scene editing flexibility while preserving rendering quality and spatial consistency.
October 23, 2023
Xiaoxiao Long, Yuan-Chen Guo, Cheng Lin, Yuan Liu, Zhiyang Dou, Lingjie Liu, Yuexin Ma, Song-Hai Zhang, Marc Habermann, Christian Theobalt, Wenping Wang
Wonder3D generates consistent multi-view normal maps and color images from a single input image, then fuses them into a high-quality textured 3D mesh.
October 9, 2023
Wangbo Yu, Li Yuan, Yan-Pei Cao, Xiangjun Gao, Xiaoyu Li, Wenbo Hu, Long Quan, Ying Shan, Yonghong Tian
HiFi-123 focuses on high-fidelity single-image-to-3D generation, reconstructing detailed geometry and textures through diffusion-guided multi-view consistency optimization.
News this year.
June 2026
First public preview of the world model: persistent, multi-agent, authorable from any input.
Read
May 2026
Single image to 3D, with weights and a ComfyUI node out on the same day as the paper.
Read
March 2026
One open model for automatic rigging — across arbitrary character topologies.
Read
Open research roles.
Dedicated compute, competitive pay, and papers published as they land.
Research Scientist — World Models
Beijing / ShanghaiApply
Research Scientist — 3D Generation
Beijing / Shanghai / HangzhouApply
Senior Research Scientist — 3D Vision & Graphics
Beijing / Shanghai / HangzhouApply
Research Engineer — Inference Acceleration
BeijingApply
Research Intern
Competitive daily pay · Higher for first authors at top venues
Beijing / Shanghai / HangzhouApply
