Hugging Face's logo

32V
/
MatLat

PyTorch
pbr
texture-generation
material-generation
3d
matlat
Model card Files
xet
 Community
MatLat
File size: 2,222 Bytes
d1fc522
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
---
library_name: pytorch
license: cc-by-nc-4.0
tags:
  - pbr
  - texture-generation
  - material-generation
  - 3d
  - matlat
---

# MatLat Pretrained Checkpoints

Pretrained submission-version checkpoints for
[MatLat: Material Latent Space for PBR Texture Generation](https://arxiv.org/abs/2512.17302).

- `matvae_v.ours.pt`: MatVAE material autoencoder checkpoint (`global_step=200000`).
- `matdiff_v.ours.safetensors`: MatDiff text- and geometry-conditioned diffusion checkpoint.

Use them with the [official code](https://github.com/32V/mcgen).

## Download

```bash
mkdir -p checkpoints

hf download 32V/MatLat \
  matvae_v.ours.pt \
  matdiff_v.ours.safetensors \
  --revision v1.0.0 \
  --local-dir checkpoints
```

## Inference

Run from the root of the MatLat code repository:

```bash
python inference.py \
  --config configs/submission/matdiff_v.ours.yaml \
  --weights checkpoints/matdiff_v.ours.safetensors \
  --mesh assets/motorcycle_cafe_racer/motorcycle_cafe_racer.obj \
  --prompt "a matte olive green cafe racer motorcycle with brushed steel parts" \
  --guidance 5.0 \
  --steps 30 \
  --seed 42 \
  --precision bf16 \
  --out outputs/demo
```

The MatVAE path and submission camera directions are selected by
`configs/submission/matdiff_v.ours.yaml`.

## License

The MatLat checkpoint weights are released under the
[Creative Commons Attribution-NonCommercial 4.0 International License](https://creativecommons.org/licenses/by-nc/4.0/)
(CC BY-NC 4.0). The license permits sharing and adaptation with appropriate
attribution for non-commercial purposes. It does not relicense third-party
software or model dependencies used by the official code.

## File Integrity

```text
f2c0d886823d7f11fe97220503ce08069e7c1b84915b27597a198bdce307e1e9  matvae_v.ours.pt
4f464112834f3a3d621cb780078e341026a62f700cec04ad7a20bc2d414dbb43  matdiff_v.ours.safetensors
```

## Citation

```bibtex
@inproceedings{yeo2026matlat,
  title     = {MatLat: Material Latent Space for PBR Texture Generation},
  author    = {Yeo, Kyeongmin and Min, Yunhong and Kim, Jaihoon and Sung, Minhyuk},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2026},
}
```