Dream.exe DVD LoRA Checkpoints

This repository releases LoRA checkpoints developed by the Dream.exe authors, trained on data constructed by our team for Dream.exe. These checkpoints are part of our research contribution and use DVD as the base model for video depth estimation. The corresponding benchmark data is available in the Dream.exe dataset.

These checkpoints adapt the upstream DVD: Deterministic Video Depth Estimation with Generative Priors model. They do not include the upstream DVD base checkpoint or the Wan2.1 runtime files. Follow the Dream.exe model-assets guide to install those dependencies.

Repository structure

DVD_lora_shared/
├── model.safetensors
└── model_config.yaml

DVD_lora_specific/
└── <uid>/
    ├── model.safetensors
    └── model_config.yaml

DVD_lora_shared/ contains the shared Dream.exe LoRA. Each directory under DVD_lora_specific/ contains the task-specific LoRA selected by its Dream.exe UID. UID directory names are part of the checkpoint identity and should not be renamed.

Download the complete family into a Dream.exe checkout with:

hf download kaimingyang/DVD_for_Dream.exe \
  --local-dir checkpoints

For a single task, replace <uid> and download only its checkpoint:

hf download kaimingyang/DVD_for_Dream.exe \
  --include "DVD_lora_specific/<uid>/*" \
  --local-dir checkpoints

License

The upstream DVD project releases its model weights under CC BY-NC 4.0. These fine-tuned checkpoints are provided under the same non-commercial terms. Review the upstream DVD terms and the Dream.exe third-party notices before use.

Citation

If you use these checkpoints, please cite Dream.exe:

@article{zhao2026dreamexe,
  title   = {Dream.exe: Can Video Generation Models Dream Executable Robot Manipulation?},
  author  = {Zhao, Rui and Yang, Kaiming and Zhu, Jifeng and Chen, Siyang and Wang, Ziqi and Wu, Weijia and Lin, Kevin Qinghong and Wang, Heng and Shou, Mike Zheng},
  journal = {arXiv preprint arXiv:2606.04811},
  year    = {2026}
}

Base model acknowledgement

We thank the DVD authors for releasing the base model used in our LoRA training:

@article{zhang2026dvd,
  title   = {DVD: Deterministic Video Depth Estimation with Generative Priors},
  author  = {Zhang, Hongfei and Chen, Harold Haodong and Liao, Chenfei and He, Jing and Zhang, Zixin and Li, Haodong and Liang, Yihao and Chen, Kanghao and Ren, Bin and Zheng, Xu and Yang, Shuai and Zhou, Kun and Li, Yinchuan and Sebe, Nicu and Chen, Ying-Cong},
  journal = {arXiv preprint arXiv:2603.12250},
  year    = {2026}
}