Our take
- Activity
- 19
- Community
- 36
- Issues
- 46
- Pull requests
- 20
+239 stars in the last 30 days
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
19,480Forks
2,032Change since Nov 14, 2026: +5 stars·0 forks
Overview
LivePortrait is an implementation of a portrait animation method that transfers motion from a driving video or motion template to a source image or video. It aims to make portrait animation practical while offering controls for stitching and motion retargeting.
It is suited to developers, researchers, and creators who want to generate or edit portrait clips locally, or try the hosted demo. The project also supports animal portraits, with additional setup requirements.
Key Features
- Animate a source portrait image from a driving video.
- Use a source video for portrait video-to-video editing.
- Create and reuse motion templates in
.pklfiles. - Adjust portrait motion with stitching and retargeting controls.
- Run human and animal animation workflows, with animal mode requiring X-Pose.
- Use a Gradio interface or run inference from the command line.
- Optionally auto-crop driving videos and adjust crop parameters.
Use Cases
- Researchers can reproduce and build on the portrait animation method described in the project’s paper.
- Developers can add portrait animation to creative tools or workflows using the inference code and pretrained weights.
- Creators can animate a still portrait from a suitable driving clip for short-form video or visual experiments.
- Video editors can use a source video and driving motion for portrait video-to-video edits.
Getting Started
Clone the repository, create a Python 3.10 Conda environment, install dependencies, and download pretrained weights:
git clone https://github.com/KlingAIResearch/LivePortrait
cd LivePortrait
conda create -n LivePortrait python=3.10
conda activate LivePortrait
pip install -r requirements.txt
Then run python inference.py after downloading the weights. See the README for setup details, platform-specific instructions, and examples.
Alternatives
- MuseTalk: MuseTalk specializes in audio-driven lip synchronization, while LivePortrait animates portraits from driving video or motion templates.
| Project | Language | License | Stars | Status |
|---|---|---|---|---|
| LivePortrait | Python | Other | 19.2k | Not checked yet |
| MuseTalk | Python | Other | 6.7k | Inactive |
Considerations
- Inference requires pretrained weights, and the README lists Git, Conda, and FFmpeg as setup prerequisites.
- The README provides CUDA and PyTorch setup guidance for Linux and Windows. Apple Silicon is supported for human mode, but the README warns it may be much slower than an RTX 4090.
- Animal mode requires X-Pose and is documented as tested on Linux and Windows with an NVIDIA GPU. It is not supported on macOS.
- The optional
torch.compileacceleration is not supported on Windows or macOS. - Portrait animation can be misused to create deepfakes. The project notes that generated results can contain visual artifacts and advises responsible use.
Found this useful?
Share it with someone who would like LivePortrait.