LivePortrait is one of the most widely used open-source portrait-animation projects — a research model that takes a single still portrait and brings it to life, transferring expression and head motion from a driving video onto the face. If you keep seeing it referenced and want a plain-English answer to what LivePortrait is, how to run it, and whether it can replace hosted talking-avatar tools, this guide covers it. Source project: github.com/KwaiVGI/LivePortrait.
What is LivePortrait?
LivePortrait is an efficient portrait animation framework. Given one source image of a face and a driving signal — usually a short video of someone moving and speaking — it produces a new video where the source portrait mirrors that motion, including subtle expression, eye, and lip movement. It is known for being fast and controllable compared with heavier diffusion-based approaches.
The key thing to understand: LivePortrait is a developer tool, not a hosted website. There is no sign-up page and no browser app. You install it locally with Python and run it from the command line or a demo interface. That makes it flexible and free to run, but it also means proper setup and a capable GPU are involved.
How to run LivePortrait
LivePortrait is run locally, so you need a few prerequisites before your first animation.
Prerequisites
- A machine with an NVIDIA GPU (CPU-only is possible but slow).
- Python and a package manager such as conda.
- Git, to clone the repository and download the model weights.
Setup
- Clone the repository from the official GitHub project and install the Python dependencies.
- Download the pre-trained weights as described in the project README.
- Run inference with your source portrait and a driving video to generate the animated result, or launch the included demo interface.
Because it is a research project, expect to read the README and issues the first time through, and to manage your own environment and updates.
LivePortrait vs a hosted talking-avatar tool
LivePortrait and hosted avatar tools solve different problems.
- Choose LivePortrait if you are comfortable with Python and GPUs, you want research-level control over the animation, and you value a free, self-hosted route over speed.
- Choose a hosted tool if you just want a finished talking video fast, with no setup. VisionStory turns a single photo plus a script into a lip-synced talking avatar video right in your browser in minutes — no Python, no GPU, no driving video to record.
If your goal is a talking presenter from one photo and a script, a hosted avatar tool is usually the quicker route. See how it works in AI video creation, or make a quick talking photo to compare.
Want a talking video without installing anything? Try VisionStory free — upload a photo, type a script, and get a shareable talking avatar video in minutes.