# Zen Dub URL: https://docs.hanzo.ai/docs/projects/hanzoai/zen-dub > Zen Dub - Real-time high-quality lip-sync model > Forked from [zenlm/zen-dub](https://github.com/zenlm/zen-dub), under Apache-2.0. Upstream holds the copyright; changes here are ours. Zen-Dub: Real-Time High-Fidelity Video Dubbing via Spatio-Temporal Sampling Yue Zhang\*, Zhizhou Zhong\*, Minhao Liu\*, Zhaokang Chen, Bin Wu†, Yubin Zeng, Chao Zhan, Junxin Huang, Yingjie He, Wenjiang Zhou (\*Equal Contribution, †Corresponding Author, ) Hanzo AI Zen LM **[github](https://github.com/hanzoai/zen-dub)_\* \*_[huggingface](https://huggingface.co/hanzoai/Zen-Dub)_\* \*_[space](https://huggingface.co/spaces/hanzoai/Zen-Dub)_\* \*_[Technical report](https://arxiv.org/abs/2410.10122)** > **Attribution.** Zen-Dub is built on [MuseTalk](https://github.com/TMElyralab/MuseTalk) (MIT License, (c) 2024 Tencent Music Entertainment Lyra Lab). It also uses [sd-vae-ft-mse](https://huggingface.co/stabilityai/sd-vae-ft-mse) (Stability AI, MIT), [Whisper](https://github.com/openai/whisper) (OpenAI, MIT), and [face-alignment / 2D-FAN-4](https://github.com/1adrianb/face-alignment) (BSD-3) for the blend mask. The lip-sync UNet derives from Stable Diffusion v1.4 (CreativeML Open RAIL-M). See `NOTICE` and `THIRD-PARTY-LICENSES.md`. Zen-Dub itself is released under Apache-2.0 by the Zen Authors (). We introduce `Zen-Dub`, a **real-time high quality** lip-syncing model (30fps+ on an NVIDIA Tesla V100). Zen-Dub can be applied with input videos, e.g., generated by [MuseV](https://github.com/hanzoai/MuseV), as a complete virtual human solution. ## πŸ”₯ Updates We're excited to unveil Zen-Dub 1.5. This version \*_(1)_\* integrates training with perceptual loss, GAN loss, and sync loss, significantly boosting its overall performance. \*_(2)\*\* We've implemented a two-stage training strategy and a spatio-temporal data sampling approach to strike a balance between visual quality and lip-sync accuracy. Learn more details [here](https://arxiv.org/abs/2410.10122). \*_The inference codes, training codes and model weights of Zen-Dub 1.5 are all available now!__ πŸš€ # Overview `Zen-Dub` is a real-time high quality audio-driven lip-syncing model trained in the latent space of `ft-mse-vae`, which 1. modifies an unseen face according to the input audio, with a size of face region of `256 x 256`. 2. supports audio in various languages, such as Chinese, English, and Japanese. 3. supports real-time inference with 30fps+ on an NVIDIA Tesla V100. 4. supports modification of the center point of the face region proposes, which **SIGNIFICANTLY** affects generation results. 5. checkpoint available trained on the HDTF and private dataset. # News - \[04/05/2025] :mega: We are excited to announce that the training code is now open-sourced! You can now train your own Zen-Dub model using our provided training scripts and configurations. - \[03/28/2025] We are thrilled to announce the release of our 1.5 version. This version is a significant improvement over the 1.0 version, with enhanced clarity, identity consistency, and precise lip-speech synchronization. We update the [technical report](https://arxiv.org/abs/2410.10122) with more details. - \[10/18/2024] We release the [technical report](https://arxiv.org/abs/2410.10122v2). Our report details a superior model to the open-source L1 loss version. It includes GAN and perceptual losses for improved clarity, and sync loss for enhanced performance. - \[04/17/2024] We release a pipeline that utilizes Zen-Dub for real-time inference. - \[04/16/2024] Release Gradio [demo](https://huggingface.co/spaces/hanzoai/Zen-Dub) on HuggingFace Spaces (thanks to HF team for their community grant) - \[04/02/2024] Release Zen-Dub project and pretrained models. ## Model ![Model Structure](https://github.com/user-attachments/assets/02f4a214-1bdd-4326-983c-e70b478accba) Zen-Dub was trained in latent spaces, where the images were encoded by a freezed VAE. The audio was encoded by a freezed `whisper-tiny` model. The architecture of the generation network was borrowed from the UNet of the `stable-diffusion-v1-4`, where the audio embeddings were fused to the image embeddings by cross-attention. Note that although we use a very similar architecture as Stable Diffusion, Zen-Dub is distinct in that it is **NOT** a diffusion model. Instead, Zen-Dub operates by inpainting in the latent space with a single step. ## Cases ### Input Video --- --- --- --- --- --- ### Zen-Dub 1.0 --- --- --- --- --- --- ### Zen-Dub 1.5 --- --- --- --- --- --- # TODO: - [x] trained models and inference codes. - [x] Huggingface Gradio [demo](https://huggingface.co/spaces/hanzoai/Zen-Dub). - [x] codes for real-time inference. - [x] [technical report](https://arxiv.org/abs/2410.10122v2). - [x] a better model with updated [technical report](https://arxiv.org/abs/2410.10122). - [x] realtime inference code for 1.5 version. - [x] training and data preprocessing codes. - [ ] **always** welcome to submit issues and PRs to improve this repository! 😊 # Getting Started We provide a detailed tutorial about the installation and the basic usage of Zen-Dub for new users: ## Third party integration Thanks for the third-party integration, which makes installation and use more convenient for everyone. We also hope you note that we have not verified, maintained, or updated third-party. Please refer to this project for specific results. ### [ComfyUI](https://github.com/chaojie/ComfyUI-Zen-Dub) ## Installation To prepare the Python environment and install additional packages such as opencv, diffusers, mmcv, etc., please follow the steps below: ### Build environment We recommend Python 3.10 and CUDA 11.7. Set up your environment as follows: ```shell conda create -n Zen-Dub python==3.10 conda activate Zen-Dub ``` ### Install PyTorch 2.0.1 Choose one of the following installation methods: ```shell # Option 1: Using pip pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118 # Option 2: Using conda conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.8 -c pytorch -c nvidia ``` ### Install Dependencies Install the remaining required packages: ```shell pip install -r requirements.txt ``` ### Install MMLab Packages Install the MMLab ecosystem packages: ```bash pip install --no-cache-dir -U openmim mim install mmengine mim install "mmcv==2.0.1" mim install "mmdet==3.1.0" mim install "mmpose==1.1.0" ``` ### Setup FFmpeg 1. [Download](https://github.com/BtbN/FFmpeg-Builds/releases) the ffmpeg-static package 2. Configure FFmpeg based on your operating system: For Linux: ```bash export FFMPEG_PATH=/path/to/ffmpeg # Example: export FFMPEG_PATH=/zen-dub/ffmpeg-4.4-amd64-static ``` For Windows: Add the `ffmpeg-xxx\bin` directory to your system's PATH environment variable. Verify the installation by running `ffmpeg -version` in the command prompt - it should display the ffmpeg version information. ### Download weights You can download weights in two ways: #### Option 1: Using Download Scripts We provide two scripts for automatic downloading: For Linux: ```bash sh ./download_weights.sh ``` For Windows: ```batch # Run the script download_weights.bat ``` #### Option 2: Manual Download You can also download the weights manually from the following links: 1. Download our trained [weights](https://huggingface.co/hanzoai/Zen-Dub/tree/main) 2. Download the weights of other components: - [sd-vae-ft-mse](https://huggingface.co/stabilityai/sd-vae-ft-mse/tree/main) - [whisper](https://huggingface.co/openai/whisper-tiny/tree/main) - [dwpose](https://huggingface.co/yzd-v/DWPose/tree/main) - [syncnet](https://huggingface.co/ByteDance/LatentSync/tree/main) - [2D-FAN-4 facial landmarks](https://www.adrianbulat.com/downloads/python-fan/2DFAN4-cd938726ad.zip) (from [face-alignment](https://github.com/1adrianb/face-alignment), BSD-3) Finally, these weights should be organized in `models` as follows: ``` ./models/ β”œβ”€β”€ zen-dub β”‚ └── zen-dub.json β”‚ └── pytorch_model.bin β”œβ”€β”€ zen-dubV15 β”‚ └── zen-dub.json β”‚ └── unet.pth β”œβ”€β”€ syncnet β”‚ └── latentsync_syncnet.pt β”œβ”€β”€ dwpose β”‚ └── dw-ll_ucoco_384.pth β”œβ”€β”€ face-alignment β”‚ └── 2DFAN4-cd938726ad.pth β”œβ”€β”€ sd-vae β”‚ β”œβ”€β”€ config.json β”‚ └── diffusion_pytorch_model.bin └── whisper β”œβ”€β”€ config.json β”œβ”€β”€ pytorch_model.bin └── preprocessor_config.json ``` ## Quickstart ### Inference We provide inference scripts for both versions of Zen-Dub: #### Prerequisites Before running inference, please ensure ffmpeg is installed and accessible: ```bash # Check ffmpeg installation ffmpeg -version ``` If ffmpeg is not found, please install it first: - Windows: Download from [ffmpeg-static](https://github.com/BtbN/FFmpeg-Builds/releases) and add to PATH - Linux: `sudo apt-get install ffmpeg` #### Normal Inference ##### Linux Environment ```bash # Zen-Dub 1.5 (Recommended) sh inference.sh v1.5 normal # Zen-Dub 1.0 sh inference.sh v1.0 normal ``` ##### Windows Environment Please ensure that you set the `ffmpeg_path` to match the actual location of your FFmpeg installation. ```bash # Zen-Dub 1.5 (Recommended) python -m scripts.inference --inference_config configs\inference\test.yaml --result_dir results\test --unet_model_path models\zen-dubV15\unet.pth --unet_config models\zen-dubV15\zen-dub.json --version v15 --ffmpeg_path ffmpeg-master-latest-win64-gpl-shared\bin # For Zen-Dub 1.0, change: # - models\zen-dubV15 -> models\zen-dub # - unet.pth -> pytorch_model.bin # - --version v15 -> --version v1 ``` #### Real-time Inference ##### Linux Environment ```bash # Zen-Dub 1.5 (Recommended) sh inference.sh v1.5 realtime # Zen-Dub 1.0 sh inference.sh v1.0 realtime ``` ##### Windows Environment ```bash # Zen-Dub 1.5 (Recommended) python -m scripts.realtime_inference --inference_config configs\inference\realtime.yaml --result_dir results\realtime --unet_model_path models\zen-dubV15\unet.pth --unet_config models\zen-dubV15\zen-dub.json --version v15 --fps 25 --ffmpeg_path ffmpeg-master-latest-win64-gpl-shared\bin # For Zen-Dub 1.0, change: # - models\zen-dubV15 -> models\zen-dub # - unet.pth -> pytorch_model.bin # - --version v15 -> --version v1 ``` The configuration file `configs/inference/test.yaml` contains the inference settings, including: - `video_path`: Path to the input video, image file, or directory of images - `audio_path`: Path to the input audio file Note: For optimal results, we recommend using input videos with 25fps, which is the same fps used during model training. If your video has a lower frame rate, you can use frame interpolation or convert it to 25fps using ffmpeg. Important notes for real-time inference: 1. Set `preparation` to `True` when processing a new avatar 2. After preparation, the avatar will generate videos using audio clips from `audio_clips` 3. The generation process can achieve 30fps+ on an NVIDIA Tesla V100 4. Set `preparation` to `False` for generating more videos with the same avatar For faster generation without saving images, you can use: ```bash python -m scripts.realtime_inference --inference_config configs/inference/realtime.yaml --skip_save_images ``` ## Gradio Demo We provide an intuitive web interface through Gradio for users to easily adjust input parameters. To optimize inference time, users can generate only the **first frame** to fine-tune the best lip-sync parameters, which helps reduce facial artifacts in the final output. ![para](https://docs.hanzo.ai/docs/projects/hanzoai/assets/figs/gradio_2.png) For minimum hardware requirements, we tested the system on a Windows environment using an NVIDIA GeForce RTX 3050 Ti Laptop GPU with 4GB VRAM. In fp16 mode, generating an 8-second video takes approximately 5 minutes. ![speed](https://docs.hanzo.ai/docs/projects/hanzoai/assets/figs/gradio.png) Both Linux and Windows users can launch the demo using the following command. Please ensure that the `ffmpeg_path` parameter matches your actual FFmpeg installation path: ```bash # You can remove --use_float16 for better quality, but it will increase VRAM usage and inference time python app.py --use_float16 --ffmpeg_path ffmpeg-master-latest-win64-gpl-shared\bin ``` ## Training ### Data Preparation To train Zen-Dub, you need to prepare your dataset following these steps: 1. **Place your source videos** For example, if you're using the HDTF dataset, place all your video files in `./dataset/HDTF/source`. 2. **Run the preprocessing script** ```bash python -m scripts.preprocess --config ./configs/training/preprocess.yaml ``` This script will: - Extract frames from videos - Detect and align faces - Generate audio features - Create the necessary data structure for training ### Training Process After data preprocessing, you can start the training process: 1. **First Stage** ```bash sh train.sh stage1 ``` 2. **Second Stage** ```bash sh train.sh stage2 ``` ### Configuration Adjustment Before starting the training, you should adjust the configuration files according to your hardware and requirements: 1. **GPU Configuration** (`configs/training/gpu.yaml`): - `gpu_ids`: Specify the GPU IDs you want to use (e.g., "0,1,2,3") - `num_processes`: Set this to match the number of GPUs you're using 2. **Stage 1 Configuration** (`configs/training/stage1.yaml`): - `data.train_bs`: Adjust batch size based on your GPU memory (default: 32) - `data.n_sample_frames`: Number of sampled frames per video (default: 1) 3. **Stage 2 Configuration** (`configs/training/stage2.yaml`): - `random_init_unet`: Must be set to `False` to use the model from stage 1 - `data.train_bs`: Smaller batch size due to high GPU memory cost (default: 2) - `data.n_sample_frames`: Higher value for temporal consistency (default: 16) - `solver.gradient_accumulation_steps`: Increase to simulate larger batch sizes (default: 8) ### GPU Memory Requirements Based on our testing on a machine with 8 NVIDIA H20 GPUs: #### Stage 1 Memory Usage | Batch Size | Gradient Accumulation | Memory per GPU | Recommendation | | :--------: | :-------------------: | :------------: | :------------: | | 8 | 1 | \~32GB | | | 16 | 1 | \~45GB | | | 32 | 1 | \~74GB | βœ“ | #### Stage 2 Memory Usage | Batch Size | Gradient Accumulation | Memory per GPU | Recommendation | | :--------: | :-------------------: | :------------: | :------------: | | 1 | 8 | \~54GB | | | 2 | 2 | \~80GB | | | 2 | 8 | \~85GB | βœ“ | ## TestCases For 1.0 Image MuseV +Zen-Dub ![](https://docs.hanzo.ai/docs/projects/hanzoai/assets/demo/musk/musk.png) ![](https://docs.hanzo.ai/docs/projects/hanzoai/assets/demo/yongen/yongen.jpeg) ![](https://docs.hanzo.ai/docs/projects/hanzoai/assets/demo/sit/sit.jpeg) ![](https://docs.hanzo.ai/docs/projects/hanzoai/assets/demo/man/man.png) ![](https://docs.hanzo.ai/docs/projects/hanzoai/assets/demo/monalisa/monalisa.png) ![](https://docs.hanzo.ai/docs/projects/hanzoai/assets/demo/sun1/sun.png) ![](https://docs.hanzo.ai/docs/projects/hanzoai/assets/demo/sun2/sun.png) #### Use of bbox\_shift to have adjustable results(For 1.0) :mag\_right: We have found that upper-bound of the mask has an important impact on mouth openness. Thus, to control the mask region, we suggest using the `bbox_shift` parameter. Positive values (moving towards the lower half) increase mouth openness, while negative values (moving towards the upper half) decrease mouth openness. You can start by running with the default configuration to obtain the adjustable value range, and then re-run the script within this range. For example, in the case of `Xinying Sun`, after running the default configuration, it shows that the adjustable value rage is \[-9, 9]. Then, to decrease the mouth openness, we set the value to be `-7`. ``` python -m scripts.inference --inference_config configs/inference/test.yaml --bbox_shift -7 ``` :pushpin: More technical details can be found in [bbox\_shift](https://docs.hanzo.ai/docs/projects/hanzoai/zen-dub). #### Combining MuseV and Zen-Dub As a complete solution to virtual human generation, you are suggested to first apply [MuseV](https://github.com/hanzoai/MuseV) to generate a video (text-to-video, image-to-video or pose-to-video) by referring [this](https://github.com/hanzoai/MuseV?tab=readme-ov-file#text2video). Frame interpolation is suggested to increase frame rate. Then, you can use `Zen-Dub` to generate a lip-sync video by referring [this](https://github.com/hanzoai/Zen-Dub?tab=readme-ov-file#inference). # Limitations - Resolution: Though Zen-Dub uses a face region size of 256 x 256, which make it better than other open-source methods, it has not yet reached the theoretical resolution bound. We will continue to deal with this problem. If you need higher resolution, you could apply super resolution models in combination with Zen-Dub. - Identity preservation: Some details of the original face are not well preserved, such as mustache, lip shape and color. - Jitter: There exists some jitter as the current pipeline adopts single-frame generation. # Citation ```bib @article{zen-dub, title={Zen-Dub: Real-Time High-Fidelity Video Dubbing via Spatio-Temporal Sampling}, author={Zhang, Yue and Zhong, Zhizhou and Liu, Minhao and Chen, Zhaokang and Wu, Bin and Zeng, Yubin and Zhan, Chao and He, Yingjie and Huang, Junxin and Zhou, Wenjiang}, journal={arxiv}, year={2025} } ``` # Disclaimer/License 1. `code`: The code of Zen-Dub is released under the MIT License. There is no limitation for both academic and commercial usage. 2. `model`: The trained model are available for any purpose, even commercially. 3. `other opensource model`: Other open-source models used must comply with their license, such as `whisper`, `ft-mse-vae`, `dwpose`, `S3FD`, etc.. 4. `demo assets`: The bundled sample images, audio, and video are provided for demonstration purposes; replace them with your own media for production use. 5. `AIGC`: This project strives to impact the domain of AI-driven video generation positively. Users are granted the freedom to create videos using this tool, but they are expected to comply with local laws and utilize it responsibly. The developers do not assume any responsibility for potential misuse by users. ## Links - [Source](https://github.com/hanzoai/zen-dub) - Licence: Apache-2.0