Datasets:

ZhengmingYu
/

DMAD-Wan-data

license: other
license_name: ultravideo-dataset-license
license_link: LICENSE
pretty_name: DMAD Wan2.1 training data
task_categories:
  - text-to-video
tags:
  - video
  - latents
  - distillation
  - webdataset
  - wan2.1
viewer: false
extra_gated_heading: Access to the DMAD Wan2.1 training data
extra_gated_description: >-
  This data is derived from the UltraVideo dataset and is released under the
  UltraVideo Dataset License Agreement (non-commercial research only).
  Requesting access means you accept the agreement below.
extra_gated_prompt: >
  # **UltraVideo Dataset License Agreement (CC-BY-4.0 with Additional
  Restrictions)**


  ### **1. Permitted Use**  

  1.1. You may **use, modify, and distribute** the **UltraVideo Dataset** (the
  "Data") for **non-commercial research purposes only**, provided you comply
  with:  
     - This Agreement,  
     - YouTube’s Terms of Service, and  
     - Any applicable copyright laws.  

  1.2. **Computational Use** (e.g., training AI models, analysis) is permitted,
  but **Results** (defined in §5) must not include more than a **de minimis
  portion** of the original Data.  


  1.3. The Data Provider waives claims against you or **Downstream Recipients**
  for compliant use, modification, or distribution.  


  ---


  ### **2. Restrictions**  

  2.1. **Commercial use is prohibited**, including:  
     - Monetization, product integration, or service improvement.  
     - Redistribution for monetary or other consideration.  

  2.2. **Ethical & Legal Compliance**:  
     - Do not use the Data for harmful, misleading, or unethical applications.  
     - Do not redistribute **raw videos, transcripts, or thumbnails** outside this dataset.  
     - Obtain **creator permission** for derivative works beyond research.  

  2.3. **No Re-identification**: Handle any residual personal data in compliance
  with privacy laws (GDPR, CCPA, etc.).  


  ---


  ### **3. Redistribution Requirements**  

  3.1. If redistributing the Data, you must:  
     - Include all original **credit/attribution** (e.g., Panda-70M/Koala-36M licenses).  
     - Bind recipients to **this Agreement**.  

  3.2. **Derivative datasets** must retain the same license terms.  


  ---


  ### **4. Disclaimers & Liability**  

  4.1. **No Ownership Warranty**: The Data Provider does not claim rights to the
  Data.  


  4.2. **"AS IS" Basis**: The Data may contain inaccuracies, biases, or
  sensitive content. Use at your own risk.  


  4.3. **No Liability**: The Data Provider and **Upstream Data Providers**
  (e.g., YouTube) are not liable for:  
     - Damages from misuse (legal, privacy, or reputational).  
     - Violations of third-party terms or laws.  

  ---


  ### **5. Definitions**  

  - **Data**: The UltraVideo Dataset (videos, metadata, etc.), modified or
  unmodified.  

  - **Results**: Outputs (e.g., AI models, research papers) containing ≤ de
  minimis Data.  

  - **Downstream Recipient**: Any entity receiving the Data through you.  


  ---


  ### **6. Termination**  

  The Data Provider may **revoke access** to the Data for legal, ethical, or
  security reasons.  


  ---


  **By using the Data, you agree to these terms.**  

  *Last updated: 2025.05.22


  --- 


  ### **Key Notes for Users**:  

  - **Non-commercial research only**.  

  - **Attribute properly** (CC-BY-4.0 + upstream licenses).  

  - **No redistribution of raw YouTube content**.  

  - **Comply with YouTube’s ToS and privacy laws**.  
extra_gated_fields:
  I will use this data for non-commercial research only and I accept the UltraVideo Dataset License Agreement: checkbox
extra_gated_button_content: I accept the license and request access

DMAD Wan2.1 training data

The training data of the DMAD 4-step students of Wan2.1-T2V (1.3B and 14B), from the paper DMAD: Distribution Matching as Adversarial Distillation for Fast Visual Generation. The training code is in train/wan of the DMAD repository; the released students are at ZhengmingYu/DMAD (wan2.1/).

This is exactly the data both students were trained on, i.e. what train/wan/experiments/dmad/prepare_data.sh produces from UltraVideo, so training can start without generating the teacher samples (Wan2.1-T2V-14B, 100 steps with classifier-free guidance, for 42,158 captions). No video is stored: the teacher samples and the real clips are kept only as Wan2.1 VAE latents.

531 GB in 5,270 WebDataset shards shards/shard_000000.tar ... shards/shard_005269.tar, 8 rows per shard (the last one 6): 42,158 rows, one per de-duplicated UltraVideo caption, row index = caption index. Each row has five members:

Member	Shape / type	What
`<index>.latent.pt`	`[16, 21, 60, 104]` bfloat16	Wan2.1-T2V-14B teacher sample for the caption: 100-step Euler, CFG 5, shift 3, seed = index, 81 frames at 832x480
`<index>.embed.pt`	`[512, 4096]` bfloat16	umT5-XXL embedding of the caption
`<index>.prompt.txt`	text	the caption (UltraVideo "Detailed Description")
`<index>.real_latent.pt`	`[16, 21, 60, 104]` bfloat16	the UltraVideo clip that the caption describes, VAE-encoded into the same latent space
`<index>.real_meta.json`	json	the clip id and its preprocessing (`mode`, `speed_factor`, `src_frames`, `src_fps`, `hdr_tonemapped`)

The real clips are preprocessed as in rcm/datasets/build_real_latents.py: 81 frames sampled uniformly over the whole clip (the teacher puts the whole caption into one clip, so the real clip is time-normalized the same way; the speed factor is recorded), resized to cover 832x480 and center-cropped, and HDR clips (about 4%) tone-mapped to SDR bt709 before encoding. Training crops both latents to the model's temporal length.

Use

hf download ZhengmingYu/DMAD-Wan-data --repo-type dataset --local-dir /path/to/data

then, in train/wan:

DATA=/path/to/data bash experiments/dmad/train_1pt3B.sh                    # 1 node x 8 GPUs
DATA=/path/to/data NODE_RANK=<i> MASTER_ADDR=<host> bash experiments/dmad/train_14B.sh   # on each of 8 nodes

The training scripts read $DATA/shards/shard*.tar with rcm/datasets/webdataset_dmad.py. To download only part of the data (e.g. for a quick test), use --include "shards/shard_0000*.tar". The latents are tied to the Wan2.1 VAE and umT5 text encoder; to rebuild or extend the data, see "Data" in the train/wan README.

License

real_latent.pt, real_meta.json and prompt.txt / embed.pt are derived from the UltraVideo dataset and are distributed under the UltraVideo Dataset License Agreement (CC-BY-4.0 with additional restrictions: non-commercial research use only, no redistribution of the source videos, derivative datasets keep these terms). By requesting access you accept that agreement; please also follow the attribution requirements of UltraVideo and its upstream sources (Panda-70M, Koala-36M).
latent.pt are AI-generated outputs of Wan2.1-T2V-14B (Apache License 2.0, see LICENSE_Wan2.1), conditioned on the UltraVideo captions.

Citation

@article{yu2026dmad,
  title={DMAD: Distribution Matching as Adversarial Distillation for Fast Visual Generation},
  author={Yu, Zhengming and Yuan, Junkun and Yang, Haotian and Qian, Gordon Guocheng and Wang, Yizhi and Wang, Angtian and Yang, Yiding and Liu, Bo and Li, Xin and Wang, Wenping and Ma, Chongyang},
  journal={arXiv preprint arXiv:2610.02188},
  year={2026}
}

@article{ultravideo,
  title={UltraVideo: High-Quality UHD Video Dataset with Comprehensive Captions},
  author={Xue, Zhucun and Zhang, Jiangning and Hu, Teng and He, Haoyang and Chen, Yinan and Cai, Yuxuan and Wang, Yabiao and Wang, Chengjie and Liu, Yong and Li, Xiangtai and Tao, Dacheng},
  journal={arXiv preprint arXiv:2506.13691},
  year={2025}
}

@article{wan2025,
  title={Wan: Open and Advanced Large-Scale Video Generative Models},
  author={Team Wan and others},
  journal={arXiv preprint arXiv:2503.20314},
  year={2025}
}

DMAD Wan2.1 training data

Contents

Use

License

Citation