\n\n# HunyuanVideo: A Systematic Framework For Large Video Generation Model Training\n\n-----\n\nThis repo contains PyTorch model definitions, pre-trained weights and inference/sampling code for our paper exploring HunyuanVideo. You can find more visualizations on our [project page](https://aivideo.hunyuan.tencent.com).\n\n> [**HunyuanVideo: A Systematic Framework For Large Video Generation Model Training**](https://arxiv.org/abs/2412.03603) \n\n\n\n## News!!\n\n* Jan 13, 2025: \ud83d\udcc8 We release the [Penguin Video Benchmark](https://github.com/Tencent/HunyuanVideo/blob/main/assets/PenguinVideoBenchmark.csv).\n* Dec 18, 2024: \ud83c\udfc3\u200d\u2642\ufe0f We release the [FP8 model weights](https://huggingface.co/tencent/HunyuanVideo/blob/main/hunyuan-video-t2v-720p/transformers/mp_rank_00_model_states_fp8.pt) of HunyuanVideo to save more GPU memory.\n* Dec 17, 2024: \ud83e\udd17 HunyuanVideo has been integrated into [Diffusers](https://huggingface.co/docs/diffusers/main/api/pipelines/hunyuan_video).\n* Dec 7, 2024: \ud83d\ude80 We release the parallel inference code for HunyuanVideo powered by [xDiT](https://github.com/xdit-project/xDiT).\n* Dec 3, 2024: \ud83d\udc4b We release the inference code and model weights of HunyuanVideo. [Download](https://github.com/Tencent/HunyuanVideo/blob/main/ckpts/README.md).\n\n\n\n## Open-source Plan\n\n- HunyuanVideo (Text-to-Video Model)\n - [x] Inference \n - [x] Checkpoints\n - [x] Multi-gpus Sequence Parallel inference (Faster inference speed on more gpus)\n - [x] Web Demo (Gradio)\n - [x] Diffusers \n - [x] FP8 Quantified weight\n - [x] Penguin Video Benchmark\n - [x] ComfyUI\n- [HunyuanVideo (Image-to-Video Model)](https://github.com/Tencent/HunyuanVideo-I2V)\n - [x] Inference \n - [x] Checkpoints \n\n\n\n## Contents\n\n- [HunyuanVideo: A Systematic Framework For Large Video Generation Model](#hunyuanvideo-a-systematic-framework-for-large-video-generation-model)\n - [News!!](#news)\n - [Open-source Plan](#open-source-plan)\n - [Contents](#contents)\n - [**Abstract**](#abstract)\n - [**HunyuanVideo Overall Architecture**](#hunyuanvideo-overall-architecture)\n - [**HunyuanVideo Key Features**](#hunyuanvideo-key-features)\n - [**Unified Image and Video Generative Architecture**](#unified-image-and-video-generative-architecture)\n - [**MLLM Text Encoder**](#mllm-text-encoder)\n - [**3D VAE**](#3d-vae)\n - [**Prompt Rewrite**](#prompt-rewrite)\n - [Comparisons](#comparisons)\n - [Requirements](#requirements)\n - [Dependencies and Installation](#\ufe0fdependencies-and-installation)\n - [Installation Guide for Linux](#installation-guide-for-linux)\n - [Download Pretrained Models](#download-pretrained-models)\n - [Single-gpu Inference](#single-gpu-inference)\n - [Using Command Line](#using-command-line)\n - [Run a Gradio Server](#run-a-gradio-server)\n - [More Configurations](#more-configurations)\n - [Parallel Inference on Multiple GPUs by xDiT](#parallel-inference-on-multiple-gpus-by-xdit)\n - [Using Command Line](#using-command-line-1)\n - [FP8 Inference](#fp8-inference)\n - [Using Command Line](#using-command-line-2)\n - [BibTeX](#bibtex)\n - [Acknowledgements](#acknowledgements)\n\n---\n\n## **Abstract**\n\nWe present HunyuanVideo, a novel open-source video foundation model that exhibits performance in video generation that is comparable to, if not superior to, leading closed-source models. In order to train HunyuanVideo model, we adopt several key technologies for model learning, including data curation, image-video joint model training, and an efficient infrastructure designed to facilitate large-scale model training and inference. Additionally, through an effective strategy for scaling model architecture and dataset, we successfully trained a video generative model with over 13 billion parameters, making it the largest among all open-source models. \n\nWe conducted extensive experiments and implemented a series of targeted designs to ensure high visual quality, motion diversity, text-video alignment, and generation stability. According to professional human evaluation results, HunyuanVideo outperforms previous state-of-the-art models, including Runway Gen-3, Luma 1.6, and 3 top-performing Chinese video generative models. By releasing the code and weights of the foundation model and its applications, we aim to bridge the gap between closed-source and open-source video foundation models. This initiative will empower everyone in the community to experiment with their ideas, fostering a more dynamic and vibrant video generation ecosystem. \n\n\n\n## **HunyuanVideo Overall Architecture**\n\nHunyuanVideo is trained on a spatial-temporally\ncompressed latent space, which is compressed through a Causal 3D VAE. Text prompts are encoded\nusing a large language model, and used as the conditions. Taking Gaussian noise and the conditions as\ninput, our generative model produces an output latent, which is then decoded to images or videos through\nthe 3D VAE decoder.\n\n
\n \n
\n\n\n\n## **HunyuanVideo Key Features**\n\n### **Unified Image and Video Generative Architecture**\n\nHunyuanVideo introduces the Transformer design and employs a Full Attention mechanism for unified image and video generation. \nSpecifically, we use a \"Dual-stream to Single-stream\" hybrid model design for video generation. In the dual-stream phase, video and text\ntokens are processed independently through multiple Transformer blocks, enabling each modality to learn its own appropriate modulation mechanisms without interference. In the single-stream phase, we concatenate the video and text\ntokens and feed them into subsequent Transformer blocks for effective multimodal information fusion.\nThis design captures complex interactions between visual and semantic information, enhancing\noverall model performance.\n\n
\n \n
\n\n\n### **MLLM Text Encoder**\n\nSome previous text-to-video models typically use pre-trained CLIP and T5-XXL as text encoders where CLIP uses Transformer Encoder and T5 uses an Encoder-Decoder structure. In contrast, we utilize a pre-trained Multimodal Large Language Model (MLLM) with a Decoder-Only structure as our text encoder, which has the following advantages: (i) Compared with T5, MLLM after visual instruction finetuning has better image-text alignment in the feature space, which alleviates the difficulty of the instruction following in diffusion models; (ii)\nCompared with CLIP, MLLM has demonstrated superior ability in image detail description\nand complex reasoning; (iii) MLLM can play as a zero-shot learner by following system instructions prepended to user prompts, helping text features pay more attention to key information. In addition, MLLM is based on causal attention while T5-XXL utilizes bidirectional attention that produces better text guidance for diffusion models. Therefore, we introduce an extra bidirectional token refiner to enhance text features.\n\n
\n \n
\n\n\n### **3D VAE**\n\nHunyuanVideo trains a 3D VAE with CausalConv3D to compress pixel-space videos and images into a compact latent space. We set the compression ratios of video length, space, and channel to 4, 8, and 16 respectively. This can significantly reduce the number of tokens for the subsequent diffusion transformer model, allowing us to train videos at the original resolution and frame rate.\n\n
\n \n
\n\n\n### **Prompt Rewrite**\n\nTo address the variability in linguistic style and length of user-provided prompts, we fine-tune the [Hunyuan-Large model](https://github.com/Tencent/Tencent-Hunyuan-Large) as our prompt rewrite model to adapt the original user prompt to model-preferred prompt.\n\nWe provide two rewrite modes: Normal mode and Master mode, which can be called using different prompts. The prompts are shown [here](hyvideo/prompt_rewrite.py). The Normal mode is designed to enhance the video generation model's comprehension of user intent, facilitating a more accurate interpretation of the instructions provided. The Master mode enhances the description of aspects such as composition, lighting, and camera movement, which leans towards generating videos with a higher visual quality. However, this emphasis may occasionally result in the loss of some semantic details. \n\nThe Prompt Rewrite Model can be directly deployed and inferred using the [Hunyuan-Large original code](https://github.com/Tencent/Tencent-Hunyuan-Large). We release the weights of the Prompt Rewrite Model [here](https://huggingface.co/Tencent/HunyuanVideo-PromptRewrite).\n\n\n\n## Comparisons\n\nTo evaluate the performance of HunyuanVideo, we selected five strong baselines from closed-source video generation models. In total, we utilized 1,533 text prompts, generating an equal number of video samples with HunyuanVideo in a single run. For a fair comparison, we conducted inference only once, avoiding any cherry-picking of results. When comparing with the baseline methods, we maintained the default settings for all selected models, ensuring consistent video resolution. Videos were assessed based on three criteria: Text Alignment, Motion Quality, and Visual Quality. More than 60 professional evaluators performed the evaluation. Notably, HunyuanVideo demonstrated the best overall performance, particularly excelling in motion quality. Please note that the evaluation is based on Hunyuan Video's high-quality version. This is different from the currently released fast version.\n\n
\n
\n \n
\n
Model
Open Source
Duration
Text Alignment
Motion Quality
Visual Quality
Overall
Ranking
\n
\n \n \n
\n
HunyuanVideo (Ours)
\u2714
5s
61.8%
66.5%
95.7%
41.3%
1
\n
\n
\n
CNTopA (API)
✘
5s
62.6%
61.7%
95.6%
37.7%
2
\n
\n
\n
CNTopB (Web)
✘
5s
60.1%
62.9%
97.7%
37.5%
3
\n
\n
\n
GEN-3 alpha (Web)
✘
6s
47.7%
54.7%
97.5%
27.4%
4
\n
\n
\n
Luma1.6 (API)
✘
5s
57.6%
44.2%
94.1%
24.8%
5
\n
\n
\n
CNTopC (Web)
✘
5s
48.4%
47.2%
96.3%
24.6%
6
\n
\n\n
\n\n\n\n\n## Requirements\n\nThe following table shows the requirements for running HunyuanVideo model (batch size = 1) to generate videos:\n\n| Model | Setting (height/width/frame) | GPU Peak Memory |\n| :----------: | :------------------------------: | :-------------: |\n| HunyuanVideo | 720px1280px129f | 60GB |\n| HunyuanVideo | 544px960px129f | 45GB |\n\n* An NVIDIA GPU with CUDA support is required. \n * The model is tested on a single 80G GPU.\n * **Minimum**: The minimum GPU memory required is 60GB for 720px1280px129f and 45G for 544px960px129f.\n * **Recommended**: We recommend using a GPU with 80GB of memory for better generation quality.\n* Tested operating system: Linux\n\n\n\n## Dependencies and Installation\n\nBegin by cloning the repository:\n\n```shell\ngit clone https://github.com/tencent/HunyuanVideo\ncd HunyuanVideo\n```\n\n### Installation Guide for Linux\n\nWe recommend CUDA versions 12.4 or 11.8 for the manual installation.\n\nConda's installation instructions are available [here](https://docs.anaconda.com/free/miniconda/index.html).\n\n```shell\n# 1. Create conda environment\nconda create -n HunyuanVideo python==3.10.9\n\n# 2. Activate the environment\nconda activate HunyuanVideo\n\n# 3. Install PyTorch and other dependencies using conda\n# For CUDA 11.8\nconda install pytorch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 pytorch-cuda=11.8 -c pytorch -c nvidia\n# For CUDA 12.4\nconda install pytorch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 pytorch-cuda=12.4 -c pytorch -c nvidia\n\n# 4. Install pip dependencies\npython -m pip install -r requirements.txt\n\n# 5. Install flash attention v2 for acceleration (requires CUDA 11.8 or above)\npython -m pip install ninja\npython -m pip install git+https://github.com/Dao-AILab/flash-attention.git@v2.6.3\n\n# 6. Install xDiT for parallel inference (It is recommended to use torch 2.4.0 and flash-attn 2.6.3)\npython -m pip install xfuser==0.4.0\n```\n\nIn case of running into float point exception(core dump) on the specific GPU type, you may try the following solutions:\n\n```shell\n# Option 1: Making sure you have installed CUDA 12.4, CUBLAS>=12.4.5.8, and CUDNN>=9.00 (or simply using our CUDA 12 docker image).\npip install nvidia-cublas-cu12==12.4.5.8\nexport LD_LIBRARY_PATH=/opt/conda/lib/python3.8/site-packages/nvidia/cublas/lib/\n\n# Option 2: Forcing to explictly use the CUDA 11.8 compiled version of Pytorch and all the other packages\npip uninstall -r requirements.txt # uninstall all packages\npip uninstall -y xfuser\npip install torch==2.4.0 --index-url https://download.pytorch.org/whl/cu118\npip install -r requirements.txt\npip install ninja\npip install git+https://github.com/Dao-AILab/flash-attention.git@v2.6.3\npip install xfuser==0.4.0\n```\n\nAdditionally, HunyuanVideo also provides a pre-built Docker image. Use the following command to pull and run the docker image.\n\n```shell\n# For CUDA 12.4 (updated to avoid float point exception)\ndocker pull hunyuanvideo/hunyuanvideo:cuda_12\ndocker run -itd --gpus all --init --net=host --uts=host --ipc=host --name hunyuanvideo --security-opt=seccomp=unconfined --ulimit=stack=67108864 --ulimit=memlock=-1 --privileged hunyuanvideo/hunyuanvideo:cuda_12\n\n# For CUDA 11.8\ndocker pull hunyuanvideo/hunyuanvideo:cuda_11\ndocker run -itd --gpus all --init --net=host --uts=host --ipc=host --name hunyuanvideo --security-opt=seccomp=unconfined --ulimit=stack=67108864 --ulimit=memlock=-1 --privileged hunyuanvideo/hunyuanvideo:cuda_11\n```\n\n\n\n## Download Pretrained Models\n\nThe details of download pretrained models are shown [here](ckpts/README.md).\n\n\n\n## Single-gpu Inference\n\nWe list the height/width/frame settings we support in the following table.\n\n| Resolution | h/w=9:16 | h/w=16:9 | h/w=4:3 | h/w=3:4 | h/w=1:1 |\n| :----------------: | :-------------: | :-------------: | :-------------: | :-------------: | :------------: |\n| 540p | 544px960px129f | 960px544px129f | 624px832px129f | 832px624px129f | 720px720px129f |\n| 720p (recommended) | 720px1280px129f | 1280px720px129f | 1104px832px129f | 832px1104px129f | 960px960px129f |\n\n### Using Command Line\n\n```bash\ncd HunyuanVideo\n\npython3 sample_video.py \\\n --video-size 720 1280 \\\n --video-length 129 \\\n --infer-steps 50 \\\n --prompt \"A cat walks on the grass, realistic style.\" \\\n --flow-reverse \\\n --use-cpu-offload \\\n --save-path ./results\n```\n\n### Run a Gradio Server\n\n```bash\npython3 gradio_server.py --flow-reverse\n\n# set SERVER_NAME and SERVER_PORT manually\n# SERVER_NAME=0.0.0.0 SERVER_PORT=8081 python3 gradio_server.py --flow-reverse\n```\n\n### More Configurations\n\nWe list some more useful configurations for easy usage:\n\n| Argument | Default | Description |\n| :--------------------: | :-------: | :----------------------------------------------------------: |\n| `--prompt` | None | The text prompt for video generation |\n| `--video-size` | 720 1280 | The size of the generated video |\n| `--video-length` | 129 | The length of the generated video |\n| `--infer-steps` | 50 | The number of steps for sampling |\n| `--embedded-cfg-scale` | 6.0 | Embedded Classifier free guidance scale |\n| `--flow-shift` | 7.0 | Shift factor for flow matching schedulers |\n| `--flow-reverse` | False | If reverse, learning/sampling from t=1 -> t=0 |\n| `--seed` | None | The random seed for generating video, if None, we init a random seed |\n| `--use-cpu-offload` | False | Use CPU offload for the model load to save more memory, necessary for high-res video generation |\n| `--save-path` | ./results | Path to save the generated video |\n\n\n\n## Parallel Inference on Multiple GPUs by xDiT\n\n[xDiT](https://github.com/xdit-project/xDiT) is a Scalable Inference Engine for Diffusion Transformers (DiTs) on multi-GPU Clusters.\nIt has successfully provided low-latency parallel inference solutions for a variety of DiTs models, including mochi-1, CogVideoX, Flux.1, SD3, etc. This repo adopted the [Unified Sequence Parallelism (USP)](https://arxiv.org/abs/2405.07719) APIs for parallel inference of the HunyuanVideo model.\n\n### Using Command Line\n\nFor example, to generate a video with 8 GPUs, you can use the following command:\n\n```bash\ncd HunyuanVideo\n\ntorchrun --nproc_per_node=8 sample_video.py \\\n --video-size 1280 720 \\\n --video-length 129 \\\n --infer-steps 50 \\\n --prompt \"A cat walks on the grass, realistic style.\" \\\n --flow-reverse \\\n --seed 42 \\\n --ulysses-degree 8 \\\n --ring-degree 1 \\\n --save-path ./results\n```\n\nYou can change the `--ulysses-degree` and `--ring-degree` to control the parallel configurations for the best performance. The valid parallel configurations are shown in the following table.\n\n\nSupported Parallel Configurations (Click to expand)\n\n\n| --video-size | --video-length | --ulysses-degree x --ring-degree | --nproc_per_node |\n| -------------------- | -------------- | -------------------------------- | ---------------- |\n| 1280 720 or 720 1280 | 129 | 8x1,4x2,2x4,1x8 | 8 |\n| 1280 720 or 720 1280 | 129 | 1x5 | 5 |\n| 1280 720 or 720 1280 | 129 | 4x1,2x2,1x4 | 4 |\n| 1280 720 or 720 1280 | 129 | 3x1,1x3 | 3 |\n| 1280 720 or 720 1280 | 129 | 2x1,1x2 | 2 |\n| 1104 832 or 832 1104 | 129 | 4x1,2x2,1x4 | 4 |\n| 1104 832 or 832 1104 | 129 | 3x1,1x3 | 3 |\n| 1104 832 or 832 1104 | 129 | 2x1,1x2 | 2 |\n| 960 960 | 129 | 6x1,3x2,2x3,1x6 | 6 |\n| 960 960 | 129 | 4x1,2x2,1x4 | 4 |\n| 960 960 | 129 | 3x1,1x3 | 3 |\n| 960 960 | 129 | 1x2,2x1 | 2 |\n| 960 544 or 544 960 | 129 | 6x1,3x2,2x3,1x6 | 6 |\n| 960 544 or 544 960 | 129 | 4x1,2x2,1x4 | 4 |\n| 960 544 or 544 960 | 129 | 3x1,1x3 | 3 |\n| 960 544 or 544 960 | 129 | 1x2,2x1 | 2 |\n| 832 624 or 624 832 | 129 | 4x1,2x2,1x4 | 4 |\n| 624 832 or 624 832 | 129 | 3x1,1x3 | 3 |\n| 832 624 or 624 832 | 129 | 2x1,1x2 | 2 |\n| 720 720 | 129 | 1x5 | 5 |\n| 720 720 | 129 | 3x1,1x3 | 3 |\n\n\n\n\n
\n
\n\n
\n
Latency (Sec) for 1280x720 (129 frames 50 steps) on 8xGPU
\n
\n
\n
1
\n
2
\n
4
\n
8
\n
\n\n\n
\n
1904.08
\n
934.09 (2.04x)
\n
514.08 (3.70x)
\n
337.58 (5.64x)
\n
\n\n\n\n
\n\n\n\n\n## FP8 Inference\n\nUsing HunyuanVideo with FP8 quantized weights, which saves about 10GB of GPU memory. You can download the [weights](https://huggingface.co/tencent/HunyuanVideo/blob/main/hunyuan-video-t2v-720p/transformers/mp_rank_00_model_states_fp8.pt) and [weight scales](https://huggingface.co/tencent/HunyuanVideo/blob/main/hunyuan-video-t2v-720p/transformers/mp_rank_00_model_states_fp8_map.pt) from Huggingface.\n\n### Using Command Line\n\nHere, you must explicitly specify the FP8 weight path. For example, to generate a video with fp8 weights, you can use the following command:\n\n```bash\ncd HunyuanVideo\n\nDIT_CKPT_PATH={PATH_TO_FP8_WEIGHTS}/{WEIGHT_NAME}_fp8.pt\n\npython3 sample_video.py \\\n --dit-weight ${DIT_CKPT_PATH} \\\n --video-size 1280 720 \\\n --video-length 129 \\\n --infer-steps 50 \\\n --prompt \"A cat walks on the grass, realistic style.\" \\\n --seed 42 \\\n --embedded-cfg-scale 6.0 \\\n --flow-shift 7.0 \\\n --flow-reverse \\\n --use-cpu-offload \\\n --use-fp8 \\\n --save-path ./results\n```\n\n\n\n## BibTeX\n\nIf you find [HunyuanVideo](https://arxiv.org/abs/2412.03603) useful for your research and applications, please cite using this BibTeX:\n\n```BibTeX\n@misc{kong2024hunyuanvideo,\n title={HunyuanVideo: A Systematic Framework For Large Video Generative Models}, \n author={Weijie Kong, Qi Tian, Zijian Zhang, Rox Min, Zuozhuo Dai, Jin Zhou, Jiangfeng Xiong, Xin Li, Bo Wu, Jianwei Zhang, Kathrina Wu, Qin Lin, Aladdin Wang, Andong Wang, Changlin Li, Duojun Huang, Fang Yang, Hao Tan, Hongmei Wang, Jacob Song, Jiawang Bai, Jianbing Wu, Jinbao Xue, Joey Wang, Junkun Yuan, Kai Wang, Mengyang Liu, Pengyu Li, Shuai Li, Weiyan Wang, Wenqing Yu, Xinchi Deng, Yang Li, Yanxin Long, Yi Chen, Yutao Cui, Yuanbo Peng, Zhentao Yu, Zhiyu He, Zhiyong Xu, Zixiang Zhou, Zunnan Xu, Yangyu Tao, Qinglin Lu, Songtao Liu, Dax Zhou, Hongfa Wang, Yong Yang, Di Wang, Yuhong Liu, and Jie Jiang, along with Caesar Zhong},\n year={2024},\n archivePrefix={arXiv preprint arXiv:2412.03603},\n primaryClass={cs.CV},\n url={https://arxiv.org/abs/2412.03603}, \n}\n```\n\n\n\n## Acknowledgements\n\nWe would like to thank the contributors to the [SD3](https://huggingface.co/stabilityai/stable-diffusion-3-medium), [FLUX](https://github.com/black-forest-labs/flux), [Llama](https://github.com/meta-llama/llama), [LLaVA](https://github.com/haotian-liu/LLaVA), [Xtuner](https://github.com/InternLM/xtuner), [diffusers](https://github.com/huggingface/diffusers) and [HuggingFace](https://huggingface.co) repositories, for their open research and exploration.\nAdditionally, we also thank the Tencent Hunyuan Multimodal team for their help with the text encoder. \n\n",
"metadata": "\"N/A\"",
"depth": 0,
"children": [
"hunyuanvideo-community/HunyuanVideo",
"Kunbyte/DRA-Ctrl",
"tencent/HunyuanCustom",
"icaruseu/QA",
"BhilVasant/Noura140",
"Usama1234/jonesjames",
"Cseti/HunyuanVideo-LoRA-Arcane_Jinx-v1",
"Cseti/HunyuanVideo-LoRA-Arcane_Style",
"1989shack/1989shack-Ecommmerce-Platform",
"jbilcke-hf/HunyuanVideo-HFIE",
"Alikhani0916/bot3",
"tanfff/test1",
"DROWHOODIS/vidgen",
"FastVideo/Hunyuan-Black-Myth-Wukong-lora-weight",
"jobs-git/HunyuanVideoCommunity",
"kudzueye/boreal-hl-v1",
"ABDALLALSWAITI/hunyuan_video_720_cfgdistill_e5m2",
"Skywork/SkyReels-V1-Hunyuan-T2V",
"Skywork/SkyReels-V1-Hunyuan-I2V",
"newgenai79/SkyReels-V1-Hunyuan-I2V-int4",
"newgenai79/HunyuanVideo-int4",
"dashtoon/hunyuan-video-keyframe-control-lora",
"neph1/AncientRome_HunyuanVideo_Lora",
"jqlive/hyv_depth_control",
"linoyts/Hunyuan-LoRA",
"Fabrice-TIERCELIN/HunyuanVideo",
"ApacheOne/Info-Hunyuan_LoRAs",
"funscripter/collection",
"Razrien/Furry-hunyuan-testing-thing"
],
"children_count": 29,
"adapters": [
"shaonroy/roy",
"trojblue/HunyuanVideo-lora-AnimeShots",
"calcuis/hyvid",
"trojblue/HunyuanVideo-lora-AnimeStills",
"a-r-r-o-w/HunyuanVideo-tuxemons",
"lucataco/hunyuan-musubi-rose-6-comfyui",
"lucataco/hunyuan-musubi-lora-heygen-6",
"lucataco/hunyuan-musubi-lora-mrgnfrmn-6",
"fofr/hunyuan-test",
"martintomov/HunyuanVideo-Coca-Cola",
"fofr/hunyuan-sonic-2",
"fofr/hunyuan-take-on-me",
"fofr/hunyuan-ponponpon",
"fofr/hunyuan-cyberpunk-mod",
"lucataco/hunyuan-lora-heygen-man-8",
"Knvl/test",
"Knvl/test2",
"Knvl/mybad",
"gj3ka1/animaengine",
"pablerdo/hunyuan-lora-f50cleat",
"lucataco/hunyuan-steamboat-willie-10",
"ghej4u/yay",
"ghej4u/ian2",
"ghej4u/lol",
"ghej4u/test",
"lucataco/hunyuan-lora-heygen-woman-2",
"deepfates/hunyuan-beast",
"hashu786/cine",
"AI-Anna/anime-renderer",
"deepfates/hunyuan-game-of-thrones",
"deepfates/hunyuan-fargo",
"deepfates/hunyuan-la-la-land",
"deepfates/hunyuan-blade-runner",
"deepfates/hunyuan-pulp-fiction",
"deepfates/hunyuan-blade-runner-2049",
"deepfates/hunyuan-the-grand-budapest-hotel",
"deepfates/hunyuan-twin-peaks",
"deepfates/hunyuan-the-neverending-story",
"deepfates/hunyuan-interstellar",
"deepfates/hunyuan-pirates-of-the-caribbean",
"deepfates/hunyuan-once-upon-a-time-in-hollywood",
"deepfates/hunyuan-dune",
"deepfates/hunyuan-arcane",
"deepfates/hunyuan-indiana-jones",
"deepfates/hunyuan-joker",
"deepfates/hunyuan-inception",
"deepfates/hunyuan-her",
"deepfates/hunyuan-westworld",
"deepfates/hunyuan-atomic-blonde",
"deepfates/hunyuan-avatar",
"deepfates/hunyuan-cowboy-bebop",
"deepfates/hunyuan-the-lord-of-the-rings",
"deepfates/hunyuan-mad-max-fury-road",
"deepfates/hunyuan-pixar",
"deepfates/hunyuan-the-matrix-trilogy",
"deepfates/hunyuan-rrr",
"deepfates/hunyuan-neon-genesis-evangelion",
"deepfates/hunyuan-spider-man-into-the-spider-verse",
"neph1/hunyuan_night_graveyard",
"CAWAI/celebdm",
"GetMonie/GawkToon",
"hashu786/CineArc",
"Alched/brxdperf_hunyuan",
"hashu786/HYVReward",
"AlekseyCalvin/hyvid_YegorLetov_concert_LoRA",
"blanflin/onstomach",
"blanflin/Standingoverfemalebj",
"CCRss/hunyuan_lora_anime_akame",
"boisterous/steak_hunyuan",
"Samsnake/LayonStomachBJ",
"Samsnake/hqgwaktoon",
"hazc138/GL1",
"istominvi/vswpntsbeige_16_16_32",
"istominvi/vswpntsbeige_30_8_32",
"istominvi/vswpntsbeige_50_8_32",
"hazc138/ZMGL",
"hazc138/gl5",
"yashlanjewar20/HEYGEN1-LORA",
"Sergidev/IllustrationTTV",
"yashlanjewar20/heygen-epoch50",
"yashlanjewar20/HeyGen-epoch16-autocaption",
"yashlanjewar20/Yash_c_epochs50_10seconds",
"yashlanjewar20/surya_10s_epoch50",
"yashlanjewar20/Yash_c_30seconds_epochs16",
"BagOu22/Lora_HKLPAZ",
"Klindle/gawk_toon3000",
"yashlanjewar20/Yash_c_16epochs_10seconds",
"yashlanjewar20/16epochs_surya_10seconds",
"ghej4u/flamingo",
"Alched/hv_dirty_panties_v1",
"ghej4u/oh",
"gulatiharsh/zinzanatrailer",
"istominvi/gocha_16_4_32",
"istominvi/gocha_32_4_32",
"istominvi/gocha_64_4_32",
"istominvi/gocha_128_4_32",
"istominvi/gocha_256_4_32",
"istominvi/gocha_16_8_32",
"istominvi/gocha_32_8_32",
"istominvi/gocha_64_8_32",
"istominvi/gocha_128_8_32",
"istominvi/gocha_256_8_32",
"StoyanG/lora-video-DrThompsonVet",
"JoshuaMKerr/joshvideo",
"trojblue/HunyuanVideo-lora-PixelArt",
"neph1/1980s_Fantasy_Movies_Hunyuan_Video_Lora",
"neph1/1920s_horror_hunyuan_video_lora",
"neph1/50s_scifi_hunyuan_video_lora",
"strangerzonehf/Hunyuan-t2v-Cartoon-LoRA"
],
"adapters_count": 109,
"quantized": [
"city96/HunyuanVideo-gguf",
"kohya-ss/HunyuanVideo-fp8_e4m3fn-unofficial"
],
"quantized_count": 2,
"merges": [],
"merges_count": 0,
"total_derivatives": 140,
"spaces": [],
"spaces_count": 0,
"parents": [],
"base_model": "tencent/HunyuanVideo",
"base_model_relation": "base"
},
{
"model_id": "hunyuanvideo-community/HunyuanVideo",
"gated": "False",
"card": "---\nbase_model:\n- tencent/HunyuanVideo\nlibrary_name: diffusers\n---\n\nUnofficial community fork for Diffusers-format weights on [`tencent/HunyuanVideo`](https://huggingface.co/tencent/HunyuanVideo).\n\n### Using Diffusers\n\nHunyuanVideo can be used directly from Diffusers. Install the latest version of Diffusers.\n\n```python\nimport torch\nfrom diffusers import HunyuanVideoPipeline, HunyuanVideoTransformer3DModel\nfrom diffusers.utils import export_to_video\n\nmodel_id = \"hunyuanvideo-community/HunyuanVideo\"\ntransformer = HunyuanVideoTransformer3DModel.from_pretrained(\n model_id, subfolder=\"transformer\", torch_dtype=torch.bfloat16\n)\npipe = HunyuanVideoPipeline.from_pretrained(model_id, transformer=transformer, torch_dtype=torch.float16)\n\n# Enable memory savings\npipe.vae.enable_tiling()\npipe.enable_model_cpu_offload()\n\noutput = pipe(\n prompt=\"A cat walks on the grass, realistic\",\n height=320,\n width=512,\n num_frames=61,\n num_inference_steps=30,\n).frames[0]\nexport_to_video(output, \"output.mp4\", fps=15)\n```\n\nRefer to the [documentation](https://huggingface.co/docs/diffusers/main/en/api/pipelines/hunyuan_video) for more information.\n\n",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [
"lucataco/hunyuan-lora-melty-test-3",
"swyne/breast-growth",
"wooyeolbaek/finetuned_models_debug2",
"wooyeolbaek/finetuned_models_videojam_debug2"
],
"adapters_count": 4,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 4,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "hunyuanvideo-community/HunyuanVideo",
"base_model_relation": "base"
},
{
"model_id": "Kunbyte/DRA-Ctrl",
"gated": "unknown",
"card": "---\nlicense: apache-2.0\nbase_model:\n- tencent/HunyuanVideo\npipeline_tag: image-to-image\nlibrary_name: diffusers\n---\n\n# Model Card for DRA-Ctrl\n\n
\n\n\n\n\n\n\n
\n\nThis repository contains the LoRA weights for DRA-Ctrl across 9 tasks. For instructions on how to use these weights, please refer to our [GitHub repository](https://github.com/Kunbyte-AI/DRA-Ctrl) and [HuggingFace Space](https://huggingface.co/spaces/Kunbyte/DRA-Ctrl).",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": null,
"base_model_relation": null
},
{
"model_id": "tencent/HunyuanCustom",
"gated": "False",
"card": "---\nlanguage:\n- en\nbase_model:\n- tencent/HunyuanVideo\npipeline_tag: image-to-video\n---\n\n\n
\n \n
\n\n# **HunyuanCustom** \ud83c\udf05\n \n
\n \n \n \n
\n
\n \n
\n
\n \n
\n-----\n\n\n> [**HunyuanCustom: A Multimodal-Driven Architecture for Customized Video Generation**](https://arxiv.org/pdf/2505.04512) \n\n\n## \ud83d\udd25\ud83d\udd25\ud83d\udd25 News!!\n* June 6, 2025: \ud83d\udc83 We release the inference code and model weights of audio-driven and video-driven powered by [OmniV2V](https://arxiv.org/abs/2506.01801).\n* May 13, 2025: \ud83c\udf89 HunyuanCustom has been integrated into [ComfyUI-HunyuanVideoWrapper](https://github.com/kijai/ComfyUI-HunyuanVideoWrapper/blob/develop/example_workflows/hyvideo_custom_testing_01.json) by [Kijai](https://github.com/kijai).\n* May 12, 2025: \ud83d\udd25 HunyuanCustom is available in Cloud-Native-Build (CNB) [HunyuanCustom](https://cnb.cool/tencent/hunyuan/HunyuanCustom).\n* May 8, 2025: \ud83d\udc4b We release the inference code and model weights of HunyuanCustom. [Download](models/README.md).\n\n\n## \ud83d\udcd1 Open-source Plan\n\n- HunyuanCustom\n - Single-Subject Video Customization\n - [x] Inference \n - [x] Checkpoints\n - [x] ComfyUI\n - Audio-Driven Video Customization\n - [x] Inference \n - [x] Checkpoints\n - [ ] ComfyUI\n - Video-Driven Video Customization\n - [x] Inference \n - [x] Checkpoints\n - [ ] ComfyUI\n - Multi-Subject Video Customization\n\n## Contents\n- [**HunyuanCustom** \ud83c\udf05](#hunyuancustom-)\n - [\ud83d\udd25\ud83d\udd25\ud83d\udd25 News!!](#-news)\n - [\ud83d\udcd1 Open-source Plan](#-open-source-plan)\n - [Contents](#contents)\n - [**Abstract**](#abstract)\n - [**HunyuanCustom Overall Architecture**](#hunyuancustom-overall-architecture)\n - [\ud83c\udf89 **HunyuanCustom Key Features**](#-hunyuancustom-key-features)\n - [**Multimodal Video customization**](#multimodal-video-customization)\n - [**Various Applications**](#various-applications)\n - [\ud83d\udcc8 Comparisons](#-comparisons)\n - [\ud83d\udcdc Requirements](#-requirements)\n - [\ud83d\udee0\ufe0f Dependencies and Installation](#\ufe0f-dependencies-and-installation)\n - [Installation Guide for Linux](#installation-guide-for-linux)\n - [\ud83e\uddf1 Download Pretrained Models](#-download-pretrained-models)\n - [\ud83d\ude80 Parallel Inference on Multiple GPUs](#-parallel-inference-on-multiple-gpus)\n - [\ud83d\udd11 Single-gpu Inference](#-single-gpu-inference)\n - [Run with very low VRAM](#run-with-very-low-vram)\n - [Run a Gradio Server](#run-a-gradio-server)\n - [\ud83d\udd17 BibTeX](#-bibtex)\n - [Acknowledgements](#acknowledgements)\n---\n\n## **Abstract**\n\nCustomized video generation aims to produce videos featuring specific subjects under flexible user-defined conditions, yet existing methods often struggle with identity consistency and limited input modalities. In this paper, we propose HunyuanCustom, a multi-modal customized video generation framework that emphasizes subject consistency while supporting image, audio, video, and text conditions. Built upon HunyuanVideo, our model first addresses the image-text conditioned generation task by introducing a text-image fusion module based on LLaVA for enhanced multi-modal understanding, along with an image ID enhancement module that leverages temporal concatenation to reinforce identity features across frames. To enable audio- and video-conditioned generation, we further propose modality-specific condition injection mechanisms: an AudioNet module that achieves hierarchical alignment via spatial cross-attention, and a video-driven injection module that integrates latent-compressed conditional video through a patchify-based feature-alignment network. Extensive experiments on single- and multi-subject scenarios demonstrate that HunyuanCustom significantly outperforms state-of-the-art open- and closed-source methods in terms of ID consistency, realism, and text-video alignment. Moreover, we validate its robustness across downstream tasks, including audio and video-driven customized video generation. Our results highlight the effectiveness of multi-modal conditioning and identity-preserving strategies in advancing controllable video generation.\n\n## **HunyuanCustom Overall Architecture**\n\n\n\nWe propose **HunyuanCustom, a multi-modal, conditional, and controllable generation model centered on subject consistency**, built upon the Hunyuan Video generation framework. It enables the generation of subject-consistent videos conditioned on text, images, audio, and video inputs. \n\n## \ud83c\udf89 **HunyuanCustom Key Features**\n\n### **Multimodal Video customization**\n\nHunyuanCustom supports inputs in the form of **text, images, audio, and video**. \nSpecifically, it can handle single or multiple image inputs to enable customized video generation for one or more subjects. \nAdditionally, it can incorporate extra audio inputs to drive the subject to speak the corresponding audio. \nLastly, HunyuanCustom supports video input, allowing for the replacement of specified objects in the video with subjects from a given image.\n\n\n### **Various Applications**\n\nWith the multi-modal capabilities of HunyuanCustom, numerous downstream tasks can be accomplished. \nFor instance, by taking multiple images as input, HunyuanCustom can facilitate **virtual human advertisements** and **virtual try-on**. Additionally, \nwith image and audio inputs, it can create **singing avatars**. Furthermore, by using an image and a video as inputs, \nHunyuanCustom supports **video editing** by replacing subjects in the video with those in the provided image. \nMore applications await your exploration!\n\n\n\n## \ud83d\udcc8 Comparisons\n\nTo evaluate the performance of HunyuanCustom, we compared it with state-of-the-art video customization methods, \nincluding VACE, Skyreels, Pika, Vidu, Keling, and Hailuo. The comparison focused on face/subject consistency, \nvideo-text alignment, and overall video quality.\n\n| Models | Face-Sim | CLIP-B-T | DINO-Sim | Temp-Consis | DD |\n|-------------------|----------|----------|----------|-------------|------|\n| VACE-1.3B | 0.204 | _0.308_ | 0.569 | **0.967** | 0.53 |\n| Skyreels | 0.402 | 0.295 | 0.579 | 0.942 | 0.72 |\n| Pika | 0.363 | 0.305 | 0.485 | 0.928 | _0.89_ |\n| Vidu2.0 | 0.424 | 0.300 | 0.537 | _0.961_ | 0.43 |\n| Keling1.6 | 0.505 | 0.285 | _0.580_ | 0.914 | 0.78 |\n| Hailuo | _0.526_ | **0.314**| 0.433 | 0.937 | **0.94** |\n| **HunyuanCustom (Ours)** | **0.627**| 0.306 | **0.593**| 0.958 | 0.71 |\n\n## \ud83d\udcdc Requirements\n\nThe following table shows the requirements for running HunyuanCustom model (batch size = 1) to generate videos:\n\n| Model | Setting (height/width/frame) | GPU Peak Memory |\n|:------------:|:--------------------------------:|:----------------:|\n| HunyuanCustom | 720px1280px129f | 80GB |\n| HunyuanCustom | 512px896px129f | 60GB |\n\n* An NVIDIA GPU with CUDA support is required. \n * The model is tested on a machine with 8GPUs.\n * **Minimum**: The minimum GPU memory required is 24GB for 720px1280px129f but very slow.\n * **Recommended**: We recommend using a GPU with 80GB of memory for better generation quality.\n* Tested operating system: Linux\n\n\n## \ud83d\udee0\ufe0f Dependencies and Installation\n\nBegin by cloning the repository:\n```shell\ngit clone https://github.com/Tencent/HunyuanCustom.git\ncd HunyuanCustom\n```\n\n### Installation Guide for Linux\n\nWe recommend CUDA versions 12.4 or 11.8 for the manual installation.\n\nConda's installation instructions are available [here](https://docs.anaconda.com/free/miniconda/index.html).\n\n```shell\n# 1. Create conda environment\nconda create -n HunyuanCustom python==3.10.9\n\n# 2. Activate the environment\nconda activate HunyuanCustom\n\n# 3. Install PyTorch and other dependencies using conda\n# For CUDA 11.8\nconda install pytorch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 pytorch-cuda=11.8 -c pytorch -c nvidia\n# For CUDA 12.4\nconda install pytorch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 pytorch-cuda=12.4 -c pytorch -c nvidia\n\n# 4. Install pip dependencies\npython -m pip install -r requirements.txt\n# 5. Install flash attention v2 for acceleration (requires CUDA 11.8 or above)\npython -m pip install ninja\npython -m pip install git+https://github.com/Dao-AILab/flash-attention.git@v2.6.3\n```\n\nIn case of running into float point exception(core dump) on the specific GPU type, you may try the following solutions:\n\n```shell\n# Option 1: Making sure you have installed CUDA 12.4, CUBLAS>=12.4.5.8, and CUDNN>=9.00 (or simply using our CUDA 12 docker image).\npip install nvidia-cublas-cu12==12.4.5.8\nexport LD_LIBRARY_PATH=/opt/conda/lib/python3.8/site-packages/nvidia/cublas/lib/\n\n# Option 2: Forcing to explicitly use the CUDA 11.8 compiled version of Pytorch and all the other packages\npip uninstall -r requirements.txt # uninstall all packages\npip install torch==2.4.0 --index-url https://download.pytorch.org/whl/cu118\npip install -r requirements.txt\npip install ninja\npip install git+https://github.com/Dao-AILab/flash-attention.git@v2.6.3\n```\n\nAdditionally, you can also use HunyuanVideo Docker image. Use the following command to pull and run the docker image.\n\n```shell\n# For CUDA 12.4 (updated to avoid float point exception)\ndocker pull hunyuanvideo/hunyuanvideo:cuda_12\ndocker run -itd --gpus all --init --net=host --uts=host --ipc=host --name hunyuanvideo --security-opt=seccomp=unconfined --ulimit=stack=67108864 --ulimit=memlock=-1 --privileged hunyuanvideo/hunyuanvideo:cuda_12\npip install gradio==3.39.0 diffusers==0.33.0 transformers==4.41.2\n\n# For CUDA 11.8\ndocker pull hunyuanvideo/hunyuanvideo:cuda_11\ndocker run -itd --gpus all --init --net=host --uts=host --ipc=host --name hunyuanvideo --security-opt=seccomp=unconfined --ulimit=stack=67108864 --ulimit=memlock=-1 --privileged hunyuanvideo/hunyuanvideo:cuda_11\npip install gradio==3.39.0 diffusers==0.33.0 transformers==4.41.2\n```\n\n\n## \ud83e\uddf1 Download Pretrained Models\n\nThe details of download pretrained models are shown [here](models/README.md).\n\n## \ud83d\ude80 Parallel Inference on Multiple GPUs\n\nFor example, to generate a video with 8 GPUs, you can use the following command:\n\n### Run Single-Subject Video Customization\n```bash\ncd HunyuanCustom\n\nexport MODEL_BASE=\"./models\"\nexport PYTHONPATH=./\ntorchrun --nnodes=1 --nproc_per_node=8 --master_port 29605 hymm_sp/sample_batch.py \\\n --ref-image './assets/images/seg_woman_01.png' \\\n --pos-prompt \"Realistic, High-quality. A woman is drinking coffee at a caf\u00e9.\" \\\n --neg-prompt \"Aerial view, aerial view, overexposed, low quality, deformation, a poor composition, bad hands, bad teeth, bad eyes, bad limbs, distortion, blurring, text, subtitles, static, picture, black border.\" \\\n --ckpt ${MODEL_BASE}\"/hunyuancustom_720P/mp_rank_00_model_states.pt\" \\\n --video-size 720 1280 \\\n --seed 1024 \\\n --sample-n-frames 129 \\\n --infer-steps 30 \\\n --flow-shift-eval-video 13.0 \\\n --save-path './results/sp_720p'\n```\n\n### Run Video-Driven Video Customization (Video Editing)\n```bash\ncd HunyuanCustom\n\nexport MODEL_BASE=\"./models\"\nexport PYTHONPATH=./\ntorchrun --nnodes=1 --nproc_per_node=8 --master_port 29605 hymm_sp/sample_batch.py \\\n --ref-image './assets/images/sed_red_panda.png' \\\n --input-video './assets/input_videos/001_bg.mp4' \\\n --mask-video './assets/input_videos/001_mask.mp4' \\\n --expand-scale 5 \\\n --video-condition \\\n --pos-prompt \"Realistic, High-quality. A red panda is walking on a stone road.\" \\\n --neg-prompt \"Aerial view, aerial view, overexposed, low quality, deformation, a poor composition, bad hands, bad teeth, bad eyes, bad limbs, distortion, blurring, text, subtitles, static, picture, black border.\" \\\n --ckpt ${MODEL_BASE}\"/hunyuancustom_editing_720P/mp_rank_00_model_states.pt\" \\\n --seed 1024 \\\n --infer-steps 50 \\\n --flow-shift-eval-video 5.0 \\\n --save-path './results/sp_editing_720p'\n # --pose-enhance # Enable for human videos to improve pose generation quality.\n```\n\n### Run Audio-Driven Video Customization\n```bash\ncd HunyuanCustom\n\nexport MODEL_BASE=\"./models\"\nexport PYTHONPATH=./\ntorchrun --nnodes=1 --nproc_per_node=8 --master_port 29605 hymm_sp/sample_batch.py \\\n --ref-image './assets/images/seg_man_01.png' \\\n --input-audio './assets/audios/milk_man.mp3' \\\n --audio-strength 0.8 \\\n --audio-condition \\\n --pos-prompt \"Realistic, High-quality. In the study, a man sits at a table featuring a bottle of milk while delivering a product presentation.\" \\\n --neg-prompt \"Two people, two persons, aerial view, overexposed, low quality, deformation, a poor composition, bad hands, bad teeth, bad eyes, bad limbs, distortion, blurring, text, subtitles, static, picture, black border.\" \\\n --ckpt ${MODEL_BASE}\"/hunyuancustom_audio_720P/mp_rank_00_model_states.pt\" \\\n --seed 1026 \\\n --video-size 720 1280 \\\n --sample-n-frames 129 \\\n --cfg-scale 7.5 \\\n --infer-steps 30 \\\n --use-deepcache 1 \\\n --flow-shift-eval-video 13.0 \\\n --save-path './results/sp_audio_720p'\n```\n\n## \ud83d\udd11 Single-gpu Inference\n\nFor example, to generate a video with 1 GPU, you can use the following command:\n\n```bash\ncd HunyuanCustom\n\nexport MODEL_BASE=\"./models\"\nexport DISABLE_SP=1\nexport PYTHONPATH=./\npython hymm_sp/sample_gpu_poor.py \\\n --ref-image './assets/images/seg_woman_01.png' \\\n --pos-prompt \"Realistic, High-quality. A woman is drinking coffee at a caf\u00e9.\" \\\n --neg-prompt \"Aerial view, aerial view, overexposed, low quality, deformation, a poor composition, bad hands, bad teeth, bad eyes, bad limbs, distortion, blurring, text, subtitles, static, picture, black border.\" \\\n --ckpt ${MODEL_BASE}\"/hunyuancustom_720P/mp_rank_00_model_states_fp8.pt\" \\\n --video-size 512 896 \\\n --seed 1024 \\\n --sample-n-frames 129 \\\n --infer-steps 30 \\\n --flow-shift-eval-video 13.0 \\\n --save-path './results/1gpu_540p' \\\n --use-fp8\n```\n\n### Run with very low VRAM\n\n```bash\ncd HunyuanCustom\n\nexport MODEL_BASE=\"./models\"\nexport CPU_OFFLOAD=1\nexport PYTHONPATH=./\npython hymm_sp/sample_gpu_poor.py \\\n --ref-image './assets/images/seg_woman_01.png' \\\n --pos-prompt \"Realistic, High-quality. A woman is drinking coffee at a caf\u00e9.\" \\\n --neg-prompt \"Aerial view, aerial view, overexposed, low quality, deformation, a poor composition, bad hands, bad teeth, bad eyes, bad limbs, distortion, blurring, text, subtitles, static, picture, black border.\" \\\n --ckpt ${MODEL_BASE}\"/hunyuancustom_720P/mp_rank_00_model_states_fp8.pt\" \\\n --video-size 720 1280 \\\n --seed 1024 \\\n --sample-n-frames 129 \\\n --infer-steps 30 \\\n --flow-shift-eval-video 13.0 \\\n --save-path './results/cpu_720p' \\\n --use-fp8 \\\n --cpu-offload \n```\n\n\n## Run a Gradio Server\n```bash\ncd HunyuanCustom\n\n# Single-Subject Video Customization\nbash ./scripts/run_gradio.sh \n\n# Video-Driven Video Customization\nbash ./scripts/run_gradio.sh --video\n\n# Audio-Driven Video Customization\nbash ./scripts/run_gradio.sh --audio\n```\n\n## \ud83d\udd17 BibTeX\n\nIf you find [HunyuanCustom](https://arxiv.org/abs/2505.04512) useful for your research and applications, please cite using this BibTeX:\n\n```BibTeX\n@misc{hu2025hunyuancustom,\n title={HunyuanCustom: A Multimodal-Driven Architecture for Customized Video Generation}, \n author={Teng Hu and Zhentao Yu and Zhengguang Zhou and Sen Liang and Yuan Zhou and Qin Lin and Qinglin Lu},\n year={2025},\n eprint={2505.04512},\n archivePrefix={arXiv},\n primaryClass={cs.CV},\n url={https://arxiv.org/abs/2505.04512}, \n}\n```\n\n## Acknowledgements\n\nWe would like to thank the contributors to the [HunyuanVideo](https://github.com/Tencent/HunyuanVideo), [HunyuanVideo-Avatar](https://github.com/Tencent-Hunyuan/HunyuanVideo-Avatar), [MimicMotion](https://github.com/Tencent/MimicMotion), [SD3](https://huggingface.co/stabilityai/stable-diffusion-3-medium), [FLUX](https://github.com/black-forest-labs/flux), [Llama](https://github.com/meta-llama/llama), [LLaVA](https://github.com/haotian-liu/LLaVA), [Xtuner](https://github.com/InternLM/xtuner), [diffusers](https://github.com/huggingface/diffusers) and [HuggingFace](https://huggingface.co) repositories, for their open research and exploration. \n",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "tencent/HunyuanCustom",
"base_model_relation": "base"
},
{
"model_id": "icaruseu/QA",
"gated": "False",
"card": "---\ndatasets:\n- fka/awesome-chatgpt-prompts\nlanguage:\n- vi\nmetrics:\n- code_eval\nbase_model:\n- tencent/HunyuanVideo\npipeline_tag: text-to-video\n---\n# Model Card for Model ID\n\n\n\nThis modelcard aims to be a base template for new models. It has been generated using [this raw template](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/modelcard_template.md?plain=1).\n\n## Model Details\n\n### Model Description\n\n\n\n\n\n- **Developed by:** [More Information Needed]\n- **Funded by [optional]:** [More Information Needed]\n- **Shared by [optional]:** [More Information Needed]\n- **Model type:** [More Information Needed]\n- **Language(s) (NLP):** [More Information Needed]\n- **License:** [More Information Needed]\n- **Finetuned from model [optional]:** [More Information Needed]\n\n### Model Sources [optional]\n\n\n\n- **Repository:** [More Information Needed]\n- **Paper [optional]:** [More Information Needed]\n- **Demo [optional]:** [More Information Needed]\n\n## Uses\n\n\n\n### Direct Use\n\n\n\n[More Information Needed]\n\n### Downstream Use [optional]\n\n\n\n[More Information Needed]\n\n### Out-of-Scope Use\n\n\n\n[More Information Needed]\n\n## Bias, Risks, and Limitations\n\n\n\n[More Information Needed]\n\n### Recommendations\n\n\n\nUsers (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.\n\n## How to Get Started with the Model\n\nUse the code below to get started with the model.\n\n[More Information Needed]\n\n## Training Details\n\n### Training Data\n\n\n\n[More Information Needed]\n\n### Training Procedure\n\n\n\n#### Preprocessing [optional]\n\n[More Information Needed]\n\n\n#### Training Hyperparameters\n\n- **Training regime:** [More Information Needed] \n\n#### Speeds, Sizes, Times [optional]\n\n\n\n[More Information Needed]\n\n## Evaluation\n\n\n\n### Testing Data, Factors & Metrics\n\n#### Testing Data\n\n\n\n[More Information Needed]\n\n#### Factors\n\n\n\n[More Information Needed]\n\n#### Metrics\n\n\n\n[More Information Needed]\n\n### Results\n\n[More Information Needed]\n\n#### Summary\n\n\n\n## Model Examination [optional]\n\n\n\n[More Information Needed]\n\n## Environmental Impact\n\n\n\nCarbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).\n\n- **Hardware Type:** [More Information Needed]\n- **Hours used:** [More Information Needed]\n- **Cloud Provider:** [More Information Needed]\n- **Compute Region:** [More Information Needed]\n- **Carbon Emitted:** [More Information Needed]\n\n## Technical Specifications [optional]\n\n### Model Architecture and Objective\n\n[More Information Needed]\n\n### Compute Infrastructure\n\n[More Information Needed]\n\n#### Hardware\n\n[More Information Needed]\n\n#### Software\n\n[More Information Needed]\n\n## Citation [optional]\n\n\n\n**BibTeX:**\n\n[More Information Needed]\n\n**APA:**\n\n[More Information Needed]\n\n## Glossary [optional]\n\n\n\n[More Information Needed]\n\n## More Information [optional]\n\n[More Information Needed]\n\n## Model Card Authors [optional]\n\n[More Information Needed]\n\n## Model Card Contact\n\n[More Information Needed]",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "icaruseu/QA",
"base_model_relation": "base"
},
{
"model_id": "BhilVasant/Noura140",
"gated": "False",
"card": "---\nlanguage:\n- hi\n- en\n- gu\nbase_model:\n- tencent/HunyuanVideo\n- tencent/HunyuanVideo-PromptRewrite\n---",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "BhilVasant/Noura140",
"base_model_relation": "base"
},
{
"model_id": "Usama1234/jonesjames",
"gated": "False",
"card": "---\nlicense: openrail\ndatasets:\n- microsoft/orca-agentinstruct-1M-v1\nlanguage:\n- ae\nmetrics:\n- accuracy\nbase_model:\n- tencent/HunyuanVideo\nnew_version: Qwen/QwQ-32B-Preview\npipeline_tag: text-classification\nlibrary_name: allennlp\ntags:\n- finance\n---",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "Usama1234/jonesjames",
"base_model_relation": "base"
},
{
"model_id": "Cseti/HunyuanVideo-LoRA-Arcane_Jinx-v1",
"gated": "False",
"card": "---\nbase_model:\n- tencent/HunyuanVideo\ntags:\n- LoRA\n- hunyuan\n---\n\n# Arcane Jinx HunyuanVideo LoRA v1\n\n
\n \n \n \n
\nPrompt: \"CSETIARCANE. A upper body shot of nfjinx, wearing an intricately detailed plate armor. The steel breastplate features ornate engravings, while articulated pauldrons protect her shoulders, their polished surface reflecting ambient light. Her blue hair stands out dramatically against the metallic armor. The armor's joints show fine craftsmanship with detailed rivets and carefully fitted plates. Small scratches and battle marks on the metal suggest authenticity and use. Her intense expression remains visible. Chainmail is visible at the gaps between plates, adding texture to the otherwise smooth metal surface.\"\n\nPrompt: \"CSETIARCANE. A full-body side view, nfjinx walking through a vast, ornate hall, her stride purposeful and measured. Her blue hair flows behind her with each step, framing her face in profile. Massive chamber around her, towering walls covered in elaborate golden patterns and archways. Thick smoke drifts lazily through the air. Wisps of the pale smoke curl around her advancing form, Her hands swing confidently at her sides\"\n\nPrompt: \"CSETIARCANE. Nfjinx with her blue hair strides through a rain-soaked cobblestone street. Her black miniskirt and white top peek out beneath a worn leather jacket, all dampened by the rain. The camera moves backward, focusing on her intense gaze. Her visible tattoos glisten with water droplets while her combat boots confidently strike the wet pavement. Behind her, neon signs in pink and blue illuminate the misty air, their glow reflecting off the wet stones and casting colorful shadows across her determined features\"\n\n\n\n## Important Notes:\n\nThis LoRA is created as part of a fan project for research purposes only and is not intended for commercial use. It is based on the TV series called Arcane which are protected by copyright. Users utilize the model at their own risk. Users are obligated to comply with copyright laws and applicable regulations. The model has been developed for non-commercial purposes, and it is not my intention to infringe on any copyright. I assume no responsibility for any damages or legal consequences arising from the use of the model. \n\n## Compatibility:\n- HunyuanVideo\n\n## ID Token / Trigger word(s):\n\nUse these in your prompt helps providing the character. See example prompt above.\n- csetiarcane, nfjinx, blue hair, black top\n\n**Please consider the following:**\n\n- If it doesn't produce the character, try increasing the lora strength to 1.2\n- With some seeds it doesn't work well, it simply doesn't produce either the style or the character\n- I definitely recommend using at least the following trigger words in your prompt: **'csetiarcane, nfjinx, blue hair'** .This is probably due to how the dataset was assembled, I'd like to fix this in later versions\"\n\n## Acknowledgment:\n\n- Thanks to the [Tencent team](https://github.com/Tencent/HunyuanVideo) for making this great model available\n- Thanks to [tdrussel](https://github.com/tdrussell) for the [diffusion-pipe](https://github.com/tdrussell/diffusion-pipe) that helps us making these LoRAs.\n- Thanks to [Kijai](https://github.com/kijai) for his great [ComfyUI integration](https://github.com/kijai/ComfyUI-HunyuanVideoWrapper)\n- Thanks to [POM](https://huggingface.co/peteromallet) for providing the computing resources. Without this, these LoRAs could not have been created.\n\n## Trainig details:\n- LR: 2e-5\n- Optimizer: adamw\n- steps: 6000\n- dataset: 40 (33x704x352) videos\n- rank: 32\n- batch size: 1\n- gradient accumulation steps: 4\n\n## Citation\n```\n@misc{kong2024hunyuanvideo,\n title={HunyuanVideo: A Systematic Framework For Large Video Generative Models}, \n author={Weijie Kong, Qi Tian, Zijian Zhang, Rox Min, Zuozhuo Dai, Jin Zhou, Jiangfeng Xiong, Xin Li, Bo Wu, Jianwei Zhang, Kathrina Wu, Qin Lin, Aladdin Wang, Andong Wang, Changlin Li, Duojun Huang, Fang Yang, Hao Tan, Hongmei Wang, Jacob Song, Jiawang Bai, Jianbing Wu, Jinbao Xue, Joey Wang, Junkun Yuan, Kai Wang, Mengyang Liu, Pengyu Li, Shuai Li, Weiyan Wang, Wenqing Yu, Xinchi Deng, Yang Li, Yanxin Long, Yi Chen, Yutao Cui, Yuanbo Peng, Zhentao Yu, Zhiyu He, Zhiyong Xu, Zixiang Zhou, Zunnan Xu, Yangyu Tao, Qinglin Lu, Songtao Liu, Daquan Zhou, Hongfa Wang, Yong Yang, Di Wang, Yuhong Liu, and Jie Jiang, along with Caesar Zhong},\n year={2024},\n archivePrefix={arXiv preprint arXiv:2412.03603},\n primaryClass={cs.CV},\n url={https://arxiv.org/abs/2412.03603}, \n}\n```",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "Cseti/HunyuanVideo-LoRA-Arcane_Jinx-v1",
"base_model_relation": "base"
},
{
"model_id": "Cseti/HunyuanVideo-LoRA-Arcane_Style",
"gated": "False",
"card": "---\nbase_model:\n- tencent/HunyuanVideo\ntags:\n- LoRA\n- hunyuan\n---\n\n# v3 Arcane Style HunyuanVideo LoRA is released\n
\n \n \n \n
\n\n## ID Token / Trigger word(s):\nUse this in your prompt helps providing the style.\n\n- csetiarcane animation style\n\n## Trainig details:\nIt was trained on images and videos\n- LR: 2e-5\n- Optimizer: adamw\n- epochs: 22\n- steps: 7326\n- dataset: 135 videos and 135 images\n- repeats: 5\n- rank: 128\n- batch size: 1\n- gradient accumulation steps: 4\n\n# v2 Arcane Style HunyuanVideo LoRA is released\n\n## Advantages:\n- Better image quality than the v1\n\n## ID Token / Trigger word(s):\n\nUse this in your prompt helps providing the style.\n- csetiarcane, scene, character\n\nIf you want to generate one or more characters in Arcane style, I strongly recommend including the word 'character' in the prompt as well. This seems to be quite a strong token and helps with displaying the style.\n\n## Trainig details:\n- LR: 2e-5\n- Optimizer: adamw\n- epochs: 52\n- dataset: 120 (33x704x352) videos\n- repeats: 5\n- rank: 32\n- batch size: 1\n- gradient accumulation steps: 4\n\n# Arcane Style HunyuanVideo LoRA v1\n\n
\n \n \n \n
\n\n## Important Notes:\n\nThis LoRA is created as part of a fan project for research purposes only and is not intended for commercial use. It is based on the TV series called Arcane which are protected by copyright. Users utilize the model at their own risk. Users are obligated to comply with copyright laws and applicable regulations. The model has been developed for non-commercial purposes, and it is not my intention to infringe on any copyright. I assume no responsibility for any damages or legal consequences arising from the use of the model. \n\n## Compatibility:\n- HunyuanVideo\n\n## ID Token / Trigger word(s):\n\nUse this in your prompt helps providing the style.\n- csetiarcane\n\n**Please consider the following:**\n\n- With some seeds it doesn't work well, it simply doesn't produce either the style\n\n## Acknowledgment:\n\n- Thanks to the [Tencent team](https://github.com/Tencent/HunyuanVideo) for making this great model available\n- Thanks to [tdrussel](https://github.com/tdrussell) for the [diffusion-pipe](https://github.com/tdrussell/diffusion-pipe) that helps us making these LoRAs.\n- Thanks to [Kijai](https://github.com/kijai) for his great [ComfyUI integration](https://github.com/kijai/ComfyUI-HunyuanVideoWrapper)\n- Thanks to [POM](https://huggingface.co/peteromallet) for providing the computing resources. Without this, these LoRAs could not have been created.\n\n## Trainig details:\n- LR: 1e-4\n- Optimizer: adamw\n- epochs: 14\n- dataset: 135 (33x704x352) videos\n- repeats: 5\n- rank: 32\n- batch size: 1\n- gradient accumulation steps: 4\n\n## Citation\n```\n@misc{kong2024hunyuanvideo,\n title={HunyuanVideo: A Systematic Framework For Large Video Generative Models}, \n author={Weijie Kong, Qi Tian, Zijian Zhang, Rox Min, Zuozhuo Dai, Jin Zhou, Jiangfeng Xiong, Xin Li, Bo Wu, Jianwei Zhang, Kathrina Wu, Qin Lin, Aladdin Wang, Andong Wang, Changlin Li, Duojun Huang, Fang Yang, Hao Tan, Hongmei Wang, Jacob Song, Jiawang Bai, Jianbing Wu, Jinbao Xue, Joey Wang, Junkun Yuan, Kai Wang, Mengyang Liu, Pengyu Li, Shuai Li, Weiyan Wang, Wenqing Yu, Xinchi Deng, Yang Li, Yanxin Long, Yi Chen, Yutao Cui, Yuanbo Peng, Zhentao Yu, Zhiyu He, Zhiyong Xu, Zixiang Zhou, Zunnan Xu, Yangyu Tao, Qinglin Lu, Songtao Liu, Daquan Zhou, Hongfa Wang, Yong Yang, Di Wang, Yuhong Liu, and Jie Jiang, along with Caesar Zhong},\n year={2024},\n archivePrefix={arXiv preprint arXiv:2412.03603},\n primaryClass={cs.CV},\n url={https://arxiv.org/abs/2412.03603}, \n}\n```",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "Cseti/HunyuanVideo-LoRA-Arcane_Style",
"base_model_relation": "base"
},
{
"model_id": "1989shack/1989shack-Ecommmerce-Platform",
"gated": "False",
"card": "---\nlicense: pddl\ndatasets:\n- fka/awesome-chatgpt-prompts\nlanguage:\n- ab\nmetrics:\n- bleu\nbase_model:\n- tencent/HunyuanVideo\nnew_version: Qwen/QwQ-32B-Preview\nlibrary_name: pyannote-audio\n---",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "1989shack/1989shack-Ecommmerce-Platform",
"base_model_relation": "base"
},
{
"model_id": "jbilcke-hf/HunyuanVideo-HFIE",
"gated": "False",
"card": "---\nlanguage:\n- en\nbase_model:\n- tencent/HunyuanVideo\npipeline_tag: text-to-video\nlibrary_name: diffusers\ntags:\n- HunyuanVideo\n- Tencent\n- Video\nlicense: other\nlicense_name: tencent-hunyuan-community\nlicense_link: LICENSE\n---\n\nThis model is [HunyuanVideo](https://huggingface.co/tencent/HunyuanVideo) adapted to run on the Hugging Face Inference Endpoints.\n\n",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "jbilcke-hf/HunyuanVideo-HFIE",
"base_model_relation": "base"
},
{
"model_id": "Alikhani0916/bot3",
"gated": "unknown",
"card": "---\nlicense: afl-3.0\npipeline_tag: text-classification\nbase_model:\n- tencent/HunyuanVideo\n---",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": null,
"base_model_relation": null
},
{
"model_id": "tanfff/test1",
"gated": "False",
"card": "---\nlicense: apache-2.0\nbase_model:\n- tencent/HunyuanVideo\ntags:\n- art\n---",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "tanfff/test1",
"base_model_relation": "base"
},
{
"model_id": "DROWHOODIS/vidgen",
"gated": "False",
"card": "---\nlicense: unlicense\ndatasets:\n- HuggingFaceFW/fineweb-2\n- fka/awesome-chatgpt-prompts\n- HuggingFaceTB/finemath\n- O1-OPEN/OpenO1-SFT\n- amphora/QwQ-LongCoT-130K\n- agibot-world/AgiBotWorld-Alpha\n- CohereForAI/Global-MMLU\n- foursquare/fsq-os-places\n- deepghs/sankaku_full\n- argilla/FinePersonas-v0.1\nbase_model:\n- tencent/HunyuanVideo\nlanguage:\n- en\nmetrics:\n- accuracy\nnew_version: tencent/HunyuanVideo\npipeline_tag: text-to-video\nlibrary_name: diffusers\n---",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "DROWHOODIS/vidgen",
"base_model_relation": "base"
},
{
"model_id": "FastVideo/Hunyuan-Black-Myth-Wukong-lora-weight",
"gated": "False",
"card": "---\nbase_model:\n- tencent/HunyuanVideo\ntags:\n- LoRA\n- hunyuan\n---\n\n**Hunyuan supports Lora fine-tuning of videos up to 720p. Detailed finetuning instructions are available in our GitHub repository.** \n\nYou can easily perform inference using the LoRA weights in our [FastVideo](https://github.com/hao-ai-lab/FastVideo) repository, supporting both single and multi-GPU configurations. \nOur training dataset consists solely of Wukong videos and can be accessed [here](https://huggingface.co/datasets/FastVideo/Black-Myth-Wukong-720p)\n# Black-Myth-Wukong\n\n
\n \n
\n \n \n
\n \n
\n \n \n
\n \n
\n \n \n
\n
\nPrompt: \"Against a backdrop of ancient trees shrouded in mist, Wukong stands prominently, his sophisticated black sunglasses adding a modern edge to his mythical appearance. His face, a striking blend of human and simian traits, is characterized by intense eyes behind the dark lenses and dense fur that frames his strong features. The ornate golden armor with swirling patterns shimmers as he crosses his arms across his chest, his posture exuding authority. He nods his head rhythmically, a subtle smile playing on his lips as the sunglasses reflect the diffused light.\"\n\nPrompt: \"Through tranquil space with traditional decorations, Wukong holds red envelopes, his stylish sunglasses creating an intriguing blend with his fur-covered face showing generous spirit. His elaborate golden armor adorned with intricate patterns gleams beside lucky packets, his strong features expressing giving joy.\"\n\nPrompt: \"Against peaceful light, Wukong examines a bespoke leather journal, his black sunglasses framing his fur-covered face thoughtfully. His elaborate golden armor with intricate patterns gleams as he appreciates craftmanship, his strong simian features showing writer's interest.\"\n\nPrompt: \"In misty light among paper-cut designs, Wukong makes a respectful gesture, his sleek sunglasses harmonizing with his fur-covered face showing artistic appreciation. His elaborate golden armor with dragon patterns catches intricate shadows as he shares cultural greetings, his strong simian features radiating tradition.\"\n\nPrompt: \"In misty light, Wukong contemplates a chessboard, his fur-covered face showing thoughtful consideration. His elaborate golden armor with intricate patterns gleams as he studies the pieces, his strong features deep in strategic thought.\"\n\nPrompt: \"Against a peaceful backdrop decorated with paper-cut designs, Wukong stands with a tray of mandarin oranges, his stylish sunglasses harmonizing with his fur-covered face showing gracious hospitality. His elaborate golden armor adorned with swirling patterns catches the gentle light as he offers the lucky fruit, his strong simian features radiating traditional courtesy.\"\n\n\n## Trainig details:\n- LR: 1e-4\n- Optimizer: adamw\n- steps: 6000\n- dataset: 22 (1280x720) videos\n- rank: 32\n- alpha: 32\n- batch size: 1\n- gradient accumulation steps: 2\n\n## Acknowledgment:\n\n- Thanks to the [Tencent team](https://github.com/Tencent/HunyuanVideo) for making this great model available\n\n## Citation\n```\n@misc{kong2024hunyuanvideo,\n title={HunyuanVideo: A Systematic Framework For Large Video Generative Models}, \n author={Weijie Kong, Qi Tian, Zijian Zhang, Rox Min, Zuozhuo Dai, Jin Zhou, Jiangfeng Xiong, Xin Li, Bo Wu, Jianwei Zhang, Kathrina Wu, Qin Lin, Aladdin Wang, Andong Wang, Changlin Li, Duojun Huang, Fang Yang, Hao Tan, Hongmei Wang, Jacob Song, Jiawang Bai, Jianbing Wu, Jinbao Xue, Joey Wang, Junkun Yuan, Kai Wang, Mengyang Liu, Pengyu Li, Shuai Li, Weiyan Wang, Wenqing Yu, Xinchi Deng, Yang Li, Yanxin Long, Yi Chen, Yutao Cui, Yuanbo Peng, Zhentao Yu, Zhiyu He, Zhiyong Xu, Zixiang Zhou, Zunnan Xu, Yangyu Tao, Qinglin Lu, Songtao Liu, Daquan Zhou, Hongfa Wang, Yong Yang, Di Wang, Yuhong Liu, and Jie Jiang, along with Caesar Zhong},\n year={2024},\n archivePrefix={arXiv preprint arXiv:2412.03603},\n primaryClass={cs.CV},\n url={https://arxiv.org/abs/2412.03603}, \n}\n```",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "FastVideo/Hunyuan-Black-Myth-Wukong-lora-weight",
"base_model_relation": "base"
},
{
"model_id": "jobs-git/HunyuanVideoCommunity",
"gated": "False",
"card": "---\nbase_model:\n- tencent/HunyuanVideo\nlibrary_name: diffusers\n---\n\nUnofficial community fork for Diffusers-format weights on [`tencent/HunyuanVideo`](https://huggingface.co/tencent/HunyuanVideo).\n\n### Using Diffusers\n\nHunyuanVideo can be used directly from Diffusers. Install the latest version of Diffusers.\n\n```python\nimport torch\nfrom diffusers import HunyuanVideoPipeline, HunyuanVideoTransformer3DModel\nfrom diffusers.utils import export_to_video\n\nmodel_id = \"hunyuanvideo-community/HunyuanVideo\"\ntransformer = HunyuanVideoTransformer3DModel.from_pretrained(\n model_id, subfolder=\"transformer\", torch_dtype=torch.bfloat16\n)\npipe = HunyuanVideoPipeline.from_pretrained(model_id, transformer=transformer, torch_dtype=torch.float16)\n\n# Enable memory savings\npipe.vae.enable_tiling()\npipe.enable_model_cpu_offload()\n\noutput = pipe(\n prompt=\"A cat walks on the grass, realistic\",\n height=320,\n width=512,\n num_frames=61,\n num_inference_steps=30,\n).frames[0]\nexport_to_video(output, \"output.mp4\", fps=15)\n```\n\nRefer to the [documentation](https://huggingface.co/docs/diffusers/main/en/api/pipelines/hunyuan_video) for more information.\n\n",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "jobs-git/HunyuanVideoCommunity",
"base_model_relation": "base"
},
{
"model_id": "kudzueye/boreal-hl-v1",
"gated": "False",
"card": "---\nlicense: other\nlicense_name: tencent-hunyuan-community\nlicense_link: https://huggingface.co/tencent/HunyuanVideo/blob/main/LICENSE\nbase_model:\n- tencent/HunyuanVideo\npipeline_tag: text-to-video\n---\n\n# Boreal-HL\nBoring Reality Hunyuan LoRA\n\nThis LoRA is an attempt at improving on the overall detail of generations from Hunyuan. It focus in particular on things such as improved depth of field, realistic skin texture, and better lighting.\nIt works at generating both realistic short video clips and single frames for images.\n\nAt the moment, the candidate LoRA used here is overtrained. It is recommended to use a strength of around 0.6. You will also need to experiment a lot with the seeds, guidance, steps, and resolution. Try to keep steps over 35 and minimum resolution above 512x512. Guidance seems to work to varying degrees between 3.5-12.5. Higher guidance and strength may lead to more similarities in things such as scene and characters. You will want to experiment with lowering the strength while raising the resolution and guidance when you run into lots of distortion. Also swap seeds as results vary a lot.\n\n## IMPORTANT NOTES:\nThis LoRA is still very experimental and difficult to control. Even with the settings recommended above it may still be difficult to get usable results. I am continuing doing new training runs to see what will help improve it. Also trying to see if there are any new improvements with an updated workflow.\n\nIf you want an easier way of using this LoRA than with Comfyui, you might be able to try running it via [Fal's Hunyuan video LoRA](https://fal.ai/models/fal-ai/hunyuan-video-lora). Use this [LoRA URL](https://huggingface.co/kudzueye/boreal-hl-v1/resolve/main/boreal-hl-v1.safetensors?download=true) and set steps to 55 (pro mode) with 720p resolution. Expect to wait at least five minutes for it to run. Replicate should also have an option though I have not tested it out yet to verify the results.\n\n## Video Examples\n\n\n\n\n\n\n\n## Image Examples\nThe LoRA can also sometimes perform decently on images when you use the initial frame.\n\n\n\n## Training Details\nI used around 150 images only for this initial LoRA. I focused on public domain photos from around the early 2010s.\n\n- epochs = 600\n- gradient_accumulation_steps = 4\n- warmup_steps = 100\n#### Adapter\n- type = \"lora\"\n- rank = 32\n- dtype = \"bfloat16\"\n- only_double_blocks = true\n#### Optimizer\n- type = \"adamw_optimi\"\n- lr = 0.0002\n- betas = [ 0.9, 0.99,]\n- weight_decay = 0.01\n- eps = 1e-8\n\n\n## Additional Info\n\n[Full Video Demonstration](https://www.youtube.com/watch?v=0tuGBrDbXU0)\n\n[Diffusion Pipe for training](https://github.com/tdrussell/diffusion-pipe)\n\n[Gradio UI Diffusion Pipe option for training](https://github.com/alisson-anjos/diffusion-pipe-ui)\n\n",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "kudzueye/boreal-hl-v1",
"base_model_relation": "base"
},
{
"model_id": "ABDALLALSWAITI/hunyuan_video_720_cfgdistill_e5m2",
"gated": "False",
"card": "---\nlicense: apache-2.0\nbase_model:\n- tencent/HunyuanVideo\ntags:\n- wavespeed\n- quantization\n- e5m2\n- pytorch\n- video-generation\n- hunyuan\n- video\n---\nQuantized hunyuan_video model (e5m2) compatible with torch.compile and ComfyUI-WaveSpeed.",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "ABDALLALSWAITI/hunyuan_video_720_cfgdistill_e5m2",
"base_model_relation": "base"
},
{
"model_id": "Skywork/SkyReels-V1-Hunyuan-T2V",
"gated": "False",
"card": "---\nlicense: apache-2.0\nlanguage:\n- en\nbase_model:\n- tencent/HunyuanVideo\npipeline_tag: text-to-video\n---\n\n# SkyReels V1: Human-Centric Video Foundation Model\n
\n\n---\nThis repo contains Diffusers-format model weights for SkyReels V1 Text-to-Video models. You can find the inference code on our github repository [SkyReels-V1](https://github.com/SkyworkAI/SkyReels-V1).\n\n## Introduction\n\nSkyReels V1 is the first and most advanced open-source human-centric video foundation model. By fine-tuning HunyuanVideo on O(10M) high-quality film and television clips, Skyreels V1 offers three key advantages:\n\n1. **Open-Source Leadership**: Our Text-to-Video model achieves state-of-the-art (SOTA) performance among open-source models, comparable to proprietary models like Kling and Hailuo.\n2. **Advanced Facial Animation**: Captures 33 distinct facial expressions with over 400 natural movement combinations, accurately reflecting human emotions.\n3. **Cinematic Lighting and Aesthetics**: Trained on high-quality Hollywood-level film and television data, each generated frame exhibits cinematic quality in composition, actor positioning, and camera angles.\n\n## \ud83d\udd11 Key Features\n\n### 1. Self-Developed Data Cleaning and Annotation Pipeline\n\nOur model is built on a self-developed data cleaning and annotation pipeline, creating a vast dataset of high-quality film, television, and documentary content.\n\n- **Expression Classification**: Categorizes human facial expressions into 33 distinct types.\n- **Character Spatial Awareness**: Utilizes 3D human reconstruction technology to understand spatial relationships between multiple people in a video, enabling film-level character positioning.\n- **Action Recognition**: Constructs over 400 action semantic units to achieve a precise understanding of human actions.\n- **Scene Understanding**: Conducts cross-modal correlation analysis of clothing, scenes, and plots.\n\n### 2. Multi-Stage Image-to-Video Pretraining\n\nOur multi-stage pretraining pipeline, inspired by the HunyuanVideo design, consists of the following stages:\n\n- **Stage 1: Model Domain Transfer Pretraining**: We use a large dataset (O(10M) of film and television content) to adapt the text-to-video model to the human-centric video domain.\n- **Stage 2: Image-to-Video Model Pretraining**: We convert the text-to-video model from Stage 1 into an image-to-video model by adjusting the conv-in parameters. This new model is then pretrained on the same dataset used in Stage 1.\n- **Stage 3: High-Quality Fine-Tuning**: We fine-tune the image-to-video model on a high-quality subset of the original dataset, ensuring superior performance and quality.\n\n## Model Introduction\n| Model Name | Resolution | Video Length | FPS | Download Link |\n|-----------------|------------|--------------|-----|---------------|\n| SkyReels-V1-Hunyuan-I2V | 544px960p | 97 | 24 | \ud83e\udd17 [Download](https://huggingface.co/Skywork/SkyReels-V1-Hunyuan-I2V) |\n| SkyReels-V1-Hunyuan-T2V (Current) | 544px960p | 97 | 24 | \ud83e\udd17 [Download](https://huggingface.co/Skywork/SkyReels-V1-Hunyuan-T2V) |\n\n## Usage\n**See the [Guide](https://github.com/SkyworkAI/SkyReels-V1) for details.**\n\n## Citation\n```BibTeX\n@misc{SkyReelsV1,\n author = {SkyReels-AI},\n title = {Skyreels V1: Human-Centric Video Foundation Model},\n year = {2025},\n publisher = {Huggingface},\n journal = {Huggingface repository},\n howpublished = {\\url{https://huggingface.co/Skywork/SkyReels-V1-Hunyuan-T2V}}\n}\n```",
"metadata": "\"N/A\"",
"depth": 1,
"children": [],
"children_count": 0,
"adapters": [],
"adapters_count": 0,
"quantized": [],
"quantized_count": 0,
"merges": [],
"merges_count": 0,
"total_derivatives": 0,
"spaces": [],
"spaces_count": 0,
"parents": [
"tencent/HunyuanVideo"
],
"base_model": "Skywork/SkyReels-V1-Hunyuan-T2V",
"base_model_relation": "base"
},
{
"model_id": "Skywork/SkyReels-V1-Hunyuan-I2V",
"gated": "False",
"card": "---\nlanguage:\n- en\nbase_model:\n- tencent/HunyuanVideo\npipeline_tag: image-to-video\n---\n\n# Skyreels V1: Human-Centric Video Foundation Model\n
![](\"https://raw.githubusercontent.com/Tencent/HunyuanVideo/refs/heads/main/assets/logo.png\"){kind=logo}
\n![](\"https://raw.githubusercontent.com/Tencent/HunyuanVideo/refs/heads/main/assets/overall.png\")
\n![](\"https://raw.githubusercontent.com/Tencent/HunyuanVideo/refs/heads/main/assets/backbone.png\")
\n![](\"https://raw.githubusercontent.com/Tencent/HunyuanVideo/refs/heads/main/assets/text_encoder.png\")
\n![](\"https://raw.githubusercontent.com/Tencent/HunyuanVideo/refs/heads/main/assets/3dvae.png\")
\n![\"arXiv\"](\"https://img.shields.io/badge/ariXv-Paper-A42C25.svg\"){kind=icon}
\n![\"arXiv\"](\"https://img.shields.io/badge/ariXv-Page-A42C25.svg\"){kind=icon}
\n![\"HuggingFace\"](\"https://img.shields.io/badge/\ud83e\udd17-Model-ffbd45.svg\"){kind=icon}
\n![\"HuggingFace\"](\"https://img.shields.io/badge/\ud83e\udd17-Space-ffbd45.svg\"){kind=icon}
\n![\"GitHub\"](\"https://img.shields.io/badge/GitHub-Code-blue.svg?logo=github&\"){kind=icon}
\n![\"Project\"](\"https://img.shields.io/badge/Project-Page-blue\"){kind=icon}
\n![](\"assets/material/logo.png\"){kind=logo}
\n![](\"https://img.shields.io/static/v1?label=HunyuanCustom%20Code&message=Github&color=blue\")
\n ![](\"https://img.shields.io/static/v1?label=Project%20Page&message=Web&color=green\")
\n ![](\"https://img.shields.io/static/v1?label=Playground&message=Web&color=green\")
\n![](\"https://img.shields.io/static/v1?label=Tech)
\n![](\"https://img.shields.io/static/v1?label=HunyuanVideo&message=HuggingFace&color=yellow\")
\n![\"Skyreels](\"assets/logo2.png\"){kind=logo}
\n