Lukeguts/LongCat-Video

1---2language:3- en4- zh5library_name: diffusers6license: mit7pipeline_tag: text-to-video8tags:9- transformers10- diffusers11- image-to-video12- video-continuation13---14 15# LongCat-Video16 17<div align="center">18  <img src="assets/longcat_logo.svg" width="45%" alt="LongCat-Video" />19</div>20<hr>21 22<div align="center" style="line-height: 1;">23  <a href='https://meituan-longcat.github.io/LongCat-Video/'><img src='https://img.shields.io/badge/Project-Page-green'></a>24  <a href='https://huggingface.co/papers/2510.22200'><img src='https://img.shields.io/badge/Paper-HuggingFace-red'></a>25  <a href='https://huggingface.co/meituan-longcat/LongCat-Video'><img src='https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-blue'></a>26</div>27 28<div align="center" style="line-height: 1;">29  <a href='https://github.com/meituan-longcat/LongCat-Flash-Chat/blob/main/figures/wechat_official_accounts.png'><img src='https://img.shields.io/badge/WeChat-LongCat-brightgreen?logo=wechat&logoColor=white'></a>  30  <a href='https://x.com/Meituan_LongCat'><img src='https://img.shields.io/badge/Twitter-LongCat-white?logo=x&logoColor=white'></a>31</div>32 33<div align="center" style="line-height: 1;">34  <a href='https://huggingface.co/meituan-longcat/LongCat-Video/blob/main/LICENSE'><img src='https://img.shields.io/badge/License-MIT-f5de53?&color=f5de53'></a>35</div>36 37## Model Introduction38We introduce LongCat-Video, a foundational video generation model with 13.6B parameters, delivering strong performance across *Text-to-Video*, *Image-to-Video*, and *Video-Continuation* generation tasks. It particularly excels in efficient and high-quality long video generation, representing our first step toward world models.39 40### Key Features41- 🌟 **Unified architecture for multiple tasks**: LongCat-Video unifies *Text-to-Video*, *Image-to-Video*, and *Video-Continuation* tasks within a single video generation framework. It natively supports all these tasks with a single model and consistently delivers strong performance across each individual task.42- 🌟 **Long video generation**: LongCat-Video is natively pretrained on *Video-Continuation* tasks, enabling it to produce minutes-long videos without color drifting or quality degradation.43- 🌟 **Efficient inference**: LongCat-Video generates $720p$, $30fps$ videos within minutes by employing a coarse-to-fine generation strategy along both the temporal and spatial axes. Block Sparse Attention further enhances efficiency, particularly at high resolutions44- 🌟 **Strong performance with multi-reward RLHF**: Powered by multi-reward Group Relative Policy Optimization (GRPO), comprehensive evaluations on both internal and public benchmarks demonstrate that LongCat-Video achieves performance comparable to leading open-source video generation models as well as the latest commercial solutions.45 46For more detail, please refer to the comprehensive [***LongCat-Video Technical Report***](https://huggingface.co/papers/2510.22200).47 48## 🎥 Teaser Video49 50<div align="center">51  <video src="https://github.com/user-attachments/assets/00fa63f0-9c4e-461a-a79e-c662ad596d7d" width="2264" height="384"> </video>52</div>53 54## Quick Start55 56### Installation57 58Clone the repo:59 60```shell61git clone https://github.com/meituan-longcat/LongCat-Video62cd LongCat-Video63```64 65Install dependencies:66 67```shell68# create conda environment69conda create -n longcat-video python=3.1070conda activate longcat-video71 72# install torch (configure according to your CUDA version)73pip install torch==2.6.0+cu124 torchvision==0.21.0+cu124 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu12474 75# install flash-attn-276pip install ninja 77pip install psutil 78pip install packaging 79pip install flash_attn==2.7.4.post180 81# install other requirements82pip install -r requirements.txt83```84 85FlashAttention-2 is enabled in the model config by default; you can also change the model config to use FlashAttention-3 or xformers.86 87### Model Download88 89| Models | Download Link |90| --- | --- |91| LongCat-Video | 🤗 [Huggingface](https://huggingface.co/meituan-longcat/LongCat-Video) |92 93Download models using huggingface-cli:94```shell95pip install "huggingface_hub[cli]"96huggingface-cli download meituan-longcat/LongCat-Video --local-dir ./weights/LongCat-Video97```98 99### Run Text-to-Video100 101```shell102# Single-GPU inference103torchrun run_demo_text_to_video.py --checkpoint_dir=./weights/LongCat-Video --enable_compile104 105# Multi-GPU inference106torchrun --nproc_per_node=2 run_demo_text_to_video.py --context_parallel_size=2 --checkpoint_dir=./weights/LongCat-Video --enable_compile107```108 109### Run Image-to-Video110 111```shell112# Single-GPU inference113torchrun run_demo_image_to_video.py --checkpoint_dir=./weights/LongCat-Video --enable_compile114 115# Multi-GPU inference116torchrun --nproc_per_node=2 run_demo_image_to_video.py --context_parallel_size=2 --checkpoint_dir=./weights/LongCat-Video --enable_compile117```118 119### Run Video-Continuation120 121```shell122# Single-GPU inference123torchrun run_demo_video_continuation.py --checkpoint_dir=./weights/LongCat-Video --enable_compile124 125# Multi-GPU inference126torchrun --nproc_per_node=2 run_demo_video_continuation.py --context_parallel_size=2 --checkpoint_dir=./weights/LongCat-Video --enable_compile127```128 129### Run Long-Video Generation130 131```shell132# Single-GPU inference133torchrun run_demo_long_video.py --checkpoint_dir=./weights/LongCat-Video --enable_compile134 135# Multi-GPU inference136torchrun --nproc_per_node=2 run_demo_long_video.py --context_parallel_size=2 --checkpoint_dir=./weights/LongCat-Video --enable_compile137```138 139### Run Interactive Video Generation140 141```shell142# Single-GPU inference143torchrun run_demo_interactive_video.py --checkpoint_dir=./weights/LongCat-Video --enable_compile144 145# Multi-GPU inference146torchrun --nproc_per_node=2 run_demo_interactive_video.py --context_parallel_size=2 --checkpoint_dir=./weights/LongCat-Video --enable_compile147```148 149### Run Streamlit150 151```shell152# Single-GPU inference153streamlit run ./run_streamlit.py --server.fileWatcherType none --server.headless=false154```155 156 157 158## Evaluation Results159 160### Text-to-Video161The *Text-to-Video* MOS evaluation results on our internal benchmark.162 163| **MOS score** | **Veo3** | **PixVerse-V5** | **Wan 2.2-T2V-A14B** | **LongCat-Video** |164|---------------|-------------------|--------------------|-------------|-------------|165| **Accessibility** | Proprietary | Proprietary | Open Source | Open Source |166| **Architecture** | - | - | MoE | Dense |167| **# Total Params** | - | - | 28B | 13.6B |168| **# Activated Params** | - | - | 14B | 13.6B |169| Text-Alignment↑ | 3.99 | 3.81 | 3.70 | 3.76 |170| Visual Quality↑ | 3.23 | 3.13 | 3.26 | 3.25 |171| Motion Quality↑ | 3.86 | 3.81 | 3.78 | 3.74 |172| Overall Quality↑ | 3.48 | 3.36 | 3.35 | 3.38 |173 174### Image-to-Video175The *Image-to-Video* MOS evaluation results on our internal benchmark.176 177| **MOS score** | **Seedance 1.0** | **Hailuo-02** | **Wan 2.2-I2V-A14B** | **LongCat-Video** |178|---------------|-------------------|--------------------|-------------|-------------|179| **Accessibility** | Proprietary | Proprietary | Open Source | Open Source |180| **Architecture** | - | - | MoE | Dense |181| **# Total Params** | - | - | 28B | 13.6B |182| **# Activated Params** | - | - | 14B | 13.6B |183| Image-Alignment↑ | 4.12 | 4.18 | 4.18 | 4.04 |184| Text-Alignment↑ | 3.70 | 3.85 | 3.33 | 3.49 |185| Visual Quality↑ | 3.22 | 3.18 | 3.23 | 3.27 |186| Motion Quality↑ | 3.77 | 3.80 | 3.79 | 3.59 |187| Overall Quality↑ | 3.35 | 3.27 | 3.26 | 3.17 |188 189## Community Works190 191Community works are welcome! Please PR or inform us in Issue to add your work.192 193- [CacheDiT](https://github.com/vipshop/cache-dit) offers Fully Cache Acceleration support for LongCat-Video with DBCache and TaylorSeer, achieved nearly 1.7x speedup without obvious loss of precision. Visit their [example](https://github.com/vipshop/cache-dit/blob/main/examples/pipeline/run_longcat_video.py) for more details.194 195 196## License Agreement197 198The **model weights** are released under the **MIT License**. 199 200Any contributions to this repository are licensed under the MIT License, unless otherwise stated. This license does not grant any rights to use Meituan trademarks or patents. 201 202See the [LICENSE](https://huggingface.co/meituan-longcat/LongCat-Video/blob/main/LICENSE) file for the full license text.203 204## Usage Considerations 205This model has not been specifically designed or comprehensively evaluated for every possible downstream application. 206 207Developers should take into account the known limitations of large language models, including performance variations across different languages, and carefully assess accuracy, safety, and fairness before deploying the model in sensitive or high-risk scenarios. 208It is the responsibility of developers and downstream users to understand and comply with all applicable laws and regulations relevant to their use case, including but not limited to data protection, privacy, and content safety requirements. 209 210Nothing in this Model Card should be interpreted as altering or restricting the terms of the MIT License under which the model is released. 211 212## Citation213We kindly encourage citation of our work if you find it useful.214 215```216@misc{meituanlongcatteam2025longcatvideotechnicalreport,217      title={LongCat-Video Technical Report}, 218      author={Meituan LongCat Team and Xunliang Cai and Qilong Huang and Zhuoliang Kang and Hongyu Li and Shijun Liang and Liya Ma and Siyu Ren and Xiaoming Wei and Rixu Xie and Tong Zhang},219      year={2025},220      eprint={2510.22200},221      archivePrefix={arXiv},222      primaryClass={cs.CV},223      url={https://arxiv.org/abs/2510.22200}, 224}225```226 227## Acknowledgements228 229We would like to thank the contributors to the [Wan](https://huggingface.co/Wan-AI), [UMT5-XXL](https://huggingface.co/google/umt5-xxl), [Diffusers](https://github.com/huggingface/diffusers) and [HuggingFace](https://huggingface.co) repositories, for their open research.230 231 232## Contact233Please contact us at <a href="mailto:longcat-team@meituan.com">longcat-team@meituan.com</a> or join our WeChat Group if you have any questions.