MatLat Pretrained Checkpoints

Pretrained submission-version checkpoints for MatLat: Material Latent Space for PBR Texture Generation.

• 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.

Download

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:

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 (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

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

Citation

@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},
}