Hallo3: Highly Dynamic and Realistic Portrait Image Animation with Diffusion Transformer Networks

Jiahao Cui¹ Hui Li¹ Yun Zhan¹ Hanlin Shang¹ Kaihui Cheng¹ Yuqi Ma¹ Shan Mu¹

Hang Zhou² Jingdong Wang² Siyu Zhu^1✉️

¹Fudan University ²Baidu Inc

I. Dataset Overview

This dataset serves as the training data for the open - source Hallo3 model, specifically created for the training of video generation models. It is dedicated to promoting the research and development in the field of video generation, providing rich and high - quality data support for relevant practitioners and researchers.

II. Data Composition

Pure Talking - Head Videos: It contains over 70 hours of pure talking - head videos, precisely focusing on the speaker's facial expressions and speech. This can provide effective data for the model to learn human language expressions and facial dynamics.

Wild - Scene Video Clips: There are more than 50 wild - scene video clips, covering a wide variety of real - world scenarios, such as bustling market streets and serene natural landscapes. This helps the model learn visual features in different scenarios.

III. Dataset Download

You can download this training dataset from HuggingFace Dataset Repo.

IV. Usage Instructions

File Extraction: After downloading, all data is compressed in the .tgz file format. You can easily extract these files to obtain the internal data.

Data Usage: This dataset is mainly used for academic research and non - commercial model training. During the use process, please ensure compliance with relevant data usage norms and ethical guidelines.

V. Dataset License

This dataset adopts the Creative Commons Attribution - NonCommercial - NoDerivatives 4.0 International License (CC BY - NC - ND 4.0). This means that you can use this dataset for non - commercial purposes with the original author and source cited, but you are not allowed to modify the data or create derivative works. Please refer to the CC BY-NC-ND 4.0 Official Documentation for specific license terms.

VI. Citation Guidelines

If you use this dataset in your research or projects, to ensure academic integrity and respect for the dataset contributors, please cite this dataset in the following format:

@misc{cui2024hallo3,
    title={Hallo3: Highly Dynamic and Realistic Portrait Image Animation with Diffusion Transformer Networks},
    author={Jiahao Cui and Hui Li and Yun Zhan and Hanlin Shang and Kaihui Cheng and Yuqi Ma and Shan Mu and Hang Zhou and Jingdong Wang and Siyu Zhu},
    year={2024},
    eprint={2412.00733},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}