MuseTalk 1.5

MuseTalk 1.5

Makes a lip-synced video from a reference image or video and a driving audio track, with a test step for the mouth area. A Gradio demo for people working with talking faces.

Made by henrybit on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: MuseTalk 1.5 is a Gradio demo of the MuseTalk model. You give it a reference image or video with a face, plus an audio track, and it produces a video where the mouth moves with the audio. A step lets you test the inpainting settings on the first frame before the full run.

The code is pinned to one commit of the MuseTalk project. The repository also holds a RESEARCH.md file with a technical assessment.

Who it suits: People who work with talking faces and want to try lip sync on their own media without building the pipeline first.

Industry | Who uses it | How they use it | Example
Media | Video editors | Match a speaker's mouth to new audio | Test a short clip with a replacement voice track
Software | Developers | Try the model before integrating it | Run the demo on one image and one audio file
Research | Computer vision researchers | Study how lip sync output behaves | Tune the mouth settings on the first frame

You can skip it if you do not have permission to use the faces or audio, since the README asks you to upload only media you may use.

Where to use it:

• A web browser, on the Space's page at Hugging Face, with nothing to install
• Windows, macOS or Linux with Python, from a copy you download

Use case | Who | What they get back
Lip sync a portrait | Creators | A video of the face speaking your audio
Dub a short clip | Editors | The clip with its mouth matched to new audio
Check the mouth area | Developers | A first-frame preview before the full run

How to set it up:

• Python 3.10 on your computer
• The Python packages listed in requirements.txt
• The system packages listed in packages.txt, which the Space installs with apt-get
• The huggingface_hub package, which gives you the hf command
• Time on the first start, since the model weights download then

1. Install the huggingface_hub package.
2. Download the Space with the hf command.
3. Move into the downloaded folder.
4. Install the pinned Gradio version.
5. Install the packages from requirements.txt.
6. Start the app with Python.

pip install -U huggingface_hub
hf download henrybit/musetalk-1-5 --repo-type space --local-dir musetalk-1-5
cd musetalk-1-5
pip install gradio==5.49.1
pip install -r requirements.txt
python app.py

How to use it: Open the Space's page, or start your own copy with the command below. Add a front-facing image or short video with a clear face, add the driving audio, run Test Inpainting, then generate the full result.

python app.py

Tool | What it does
Test Inpainting | Previews the mouth area on the first frame
bbox_shift | A setting to adjust when the mouth area looks wrong
Generate | Produces the full lip-synced video

Example: You add a front-facing photo and a short voice clip, then run Test Inpainting. If the mouth looks right, you generate the full video and get the face speaking the clip.

For newcomers and experts: If you are new to lip sync, use a clear front-facing photo and a few seconds of audio. Run the test step first, because it is quicker than a full run. For video input, the README prefers 25 fps.

If you know the field, the demo wraps MuseTalk 1.5 for ZeroGPU, where a GPU is allocated only while the generate functions run. The first start fetches several sets of weights, including the UNet and the Whisper-tiny model, which can take a few minutes. Later starts reuse the Hub cache.

You add a face image or video and audio
Test Inpainting previews the first frame
You adjust bbox_shift and margins if needed
Generate produces the lip-synced video

Good to know:

• License: the Space lists mit
• Cost to run: no secrets are required for public model downloads, and HF_TOKEN is only for cases where Hub rate limits or your setup need it
• Data it reaches: the images, video and audio you upload, plus model weights downloaded from the Hub
• Not for: media you lack permission to use, and the README asks you to disclose synthetic or altered video when you publish it

On Hugging Face, 13 November 2026

• Likes: 2
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/henrybit/musetalk-1-5
• Files: https://huggingface.co/spaces/henrybit/musetalk-1-5/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.