File size: 2,301 Bytes
96e2396
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
72f895d
96e2396
 
 
 
 
 
 
 
 
 
 
 
 
dac1a9a
 
96e2396
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
---
license: other
library_name: pytorch
tags:
  - co-speech-gesture-generation
  - co-speech-motion-generation
  - speech-driven-motion
  - diffusion-models
  - long-horizon-generation
  - beat2
  - smpl-x
  - globaldiff
datasets:
  - H-Liu1997/BEAT2
---

# GlobalDiff official checkpoints

Official checkpoints for **Mitigating Error Accumulation in Co-Speech Motion Generation via Global Rotation Diffusion and Multi-Level Constraints** (AAAI 2026), a diffusion approach to long-horizon **co-speech gesture generation**.

- [Paper](https://arxiv.org/abs/2511.10076) · [Hugging Face Paper page](https://huggingface.co/papers/2511.10076)
- [Project page](https://xiangyuezhang.com/GlobalDiff/) · [Code](https://github.com/Xiangyue-Zhang/GlobalDiff)
- [Published version](https://doi.org/10.1609/aaai.v40i15.38281)
- [Generated inference data](https://huggingface.co/datasets/X-Zhang/GlobalDiff-Inference-Data)

## Files

The checkpoint files mirror the official repository while preserving their repository-relative paths:

| Path | Role |
| --- | --- |
| `Scripts/VAE/ckpt/split/global/MaskedVAE2-HorizonFlip/best.pt` | Global rotation VAE checkpoint |
| `Scripts/VAE/ckpt/split/global/MaskedVAE3-HorizonFlip/best.pt` | Global rotation VAE checkpoint |
| `Scripts/FM/ckpt/split/SimpleSpeechModel/best.zip` | Speech model checkpoint archive |
| `Scripts/FM/EMAGE_VAE/AESKConv_240_100.bin` | Evaluation/representation asset |
See the code repository's environment checker and inference instructions for all runtime dependencies.

Verify the files from the repository root with `sha256sum -c SHA256SUMS`.

## Terms

This repository mirrors research artifacts released by the authors. No new license is granted by this model card. Code, BEAT2, SMPL-X, pretrained encoders, and other third-party assets remain subject to their respective terms.

## Citation

```bibtex
@inproceedings{zhang2026globaldiff,
  title={Mitigating Error Accumulation in Co-Speech Motion Generation via Global Rotation Diffusion and Multi-Level Constraints},
  author={Zhang, Xiangyue and Li, Jianfang and Ren, Jianqiang and Zhang, Jiaxu},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={40},
  number={15},
  pages={12834--12842},
  year={2026},
  doi={10.1609/aaai.v40i15.38281}
}
```