LivePortrait
liveportrait.github.io
- Free plan
- Windows
- macOS
- Linux
- Web
- Self-hosted
What LivePortrait is
LivePortrait is an open-source system that creates portrait video from one source image, using motion from a driving video, audio, text, or generation. It also supports portrait video editing with a source video and driving video. Stitching plus eye and lip retargeting modules provide animation controls. The project offers inference code, pretrained models, and a Gradio interface for local use, alongside an online demo through Hugging Face Spaces. Its PyTorch repository is released under the MIT License and reports generation at 12.8 ms on an RTX 4090 GPU with PyTorch. Fine-tuned models can animate cats, dogs, and pandas from human driving motion. Human mode is documented for Linux, Windows, and Apple Silicon macOS; Intel macOS is not tested. Animal mode is tested only on Linux and Windows with an NVIDIA GPU and is unsupported on macOS. Local setup requires git, conda, FFmpeg, Python 3.10, and downloaded pretrained weights.
Who it suits
LivePortrait suits developers and creators needing image-driven portrait animation or editing. It is also relevant to users who want local, self-hosted inference or an online demo.
Good to know, and what to watch
- Open-source MIT-licensed implementation
- Supports video, audio, text, or generated motion
- Includes eye and lip retargeting controls
- Offers local Gradio interface and web demo
- Can animate cats, dogs, and pandas
- Local setup requires several dependencies
- Animal mode needs Linux or Windows NVIDIA hardware
- Intel macOS is untested for human mode
- Commercial users must replace InsightFace detection models
The verdict
LivePortrait combines local inference, a web demo, and controllable portrait animation from a single image. Platform and licensing conditions matter, especially for animal workflows and commercial deployments.
What the maker publishes
- Animal animation
- An animal model can animate cats, dogs, and pandas using human-driven motion. source
- animal limitation
- Animals mode is only tested on Linux and Windows with an NVIDIA GPU and is unsupported on macOS because X-Pose does not support macOS. source
- animal support
- Fine-tuned models can animate cats, dogs, and pandas using human driving motion. source
- Animation controls
- The project includes stitching and eye and lip retargeting controls for portrait animation. source
- commercial licensing
- InsightFace models included in the project are for non-commercial research, and commercial users must replace those detection models. source
- Community integrations
- The repository lists community-developed ComfyUI nodes, a Stable Diffusion WebUI extension, and FaceFusion processors that use LivePortrait. source
- controls
- The system includes stitching and eye/lip retargeting modules for improved controllability. source
- Ethics
- The maintainers warn about deepfake misuse, note that generated results may contain visual artifacts, and disclaim legal responsibility for use of generated results. source
- Hardware limits
- The repository says animal mode is tested only on Linux and Windows with an NVIDIA GPU, and is unsupported on macOS because its X-Pose dependency does not support macOS. source
- integrations
- Community resources include FaceFusion, ComfyUI nodes, a Stable Diffusion WebUI extension, Replicate, FAL, Poe, and Hugging Face Spaces. source
- License and commercial use
- The repository includes an MIT license but says commercial users must remove and replace InsightFace detection models to comply with that license. source
- local setup
- The project requires git, conda, FFmpeg, Python 3.10, and downloaded pretrained weights. source
- Local use
- The repository provides inference code, pretrained models, and a Gradio interface for local use. source
- open source
- The official implementation is provided as a PyTorch repository and is released under the MIT License. source
- Performance
- The project reports generation speed of 12.8 ms on an RTX 4090 GPU with PyTorch. source
- platform support
- Humans mode supports Linux, Windows, and macOS with Apple Silicon; Intel macOS is not tested. source
- privacy
- Motion templates can be used to speed inference and protect privacy. source
- Privacy
- The repository says generated motion template files can speed inference and protect privacy. source
- purpose
- LivePortrait synthesizes a lifelike portrait video from a single source image, using motion from a driving video, audio, text, or generation. source
- Purpose
- LivePortrait animates a portrait from a still image using facial expressions and head pose from a driving video, audio, text, or generated motion. source
- risk notice
- The maintainers warn that portrait animation can be misused for deepfakes and state that they assume no legal responsibility for generated results. source
- Security
- The maintainers report updating the Transformers and Gradio library versions in October 2024 to avoid security vulnerabilities. source
- security updates
- The maintainers updated the transformers and gradio versions to avoid security vulnerabilities. source
- speed
- Generation speed reaches 12.8 ms on an RTX 4090 GPU with PyTorch. source
Checked against the maker's pages on 2 Oct 2026.