# LivePortrait — card for AI agents

Animates a portrait photo or video with the expressions and head motion of a driving video; also supports animals.

Category: AI presenters and avatars · maker Kuaishou (Kling team) · checked 2026-11-10

## Price
- Price: Free
- Free tier: Open weights and code, free to download; self-hosted (runs on a consumer GPU or Apple Silicon).

## License
MIT (bundled InsightFace models non-commercial) — https://github.com/KlingAIResearch/LivePortrait/blob/main/LICENSE

- commercial: **conditions** (Commercial use with conditions)
  > If you want to use the LivePortrait project for commercial purposes, you should remove and replace InsightFace’s detection models to fully comply with the MIT license.
  (https://github.com/KlingAIResearch/LivePortrait/blob/main/LICENSE)
- in app: **conditions** (Ship inside your app with conditions)
  > The models of InsightFace are for non-commercial research purposes only.
  (https://github.com/KlingAIResearch/LivePortrait/blob/main/LICENSE)
- youtube: **conditions** (Monetized YouTube with conditions)
  > If you want to use the LivePortrait project for commercial purposes, you should remove and replace InsightFace’s detection models to fully comply with the MIT license.
  (https://github.com/KlingAIResearch/LivePortrait/blob/main/LICENSE)

## Rules to follow
- LivePortrait code and weights are MIT; keep the copyright notice in copies.
- The bundled InsightFace face-detection models are non-commercial; replace them before any commercial use.

## How to connect
- Download only
- Repository: https://github.com/KlingAIResearch/LivePortrait
- Website: https://huggingface.co/KlingTeam/LivePortrait

## Notes
Hugging Face API, Oct 25, 2026: 18007 downloads, 518 likes; GitHub KlingAIResearch/LivePortrait: 19309 stars. The Hugging Face repo KwaiVGI/LivePortrait now redirects to KlingTeam/LivePortrait, and GitHub KwaiVGI/LivePortrait to KlingAIResearch/LivePortrait.

## Numbers (AgentGid data)
- Gid Score: 40
- GitHub stars: 19,530

## Lines to paste into your prompt or AGENTS.md

```
- LivePortrait (AI presenters and avatars). Animates a portrait photo or video with the expressions and head motion of a driving video; also supports animals. Price: free; commercial use with conditions; ship inside your app with conditions.
  Connect: download from https://huggingface.co/KlingTeam/LivePortrait (no API).
  Rule: LivePortrait code and weights are MIT; keep the copyright notice in copies.
  Rule: The bundled InsightFace face-detection models are non-commercial; replace them before any commercial use.
  License: https://github.com/KlingAIResearch/LivePortrait/blob/main/LICENSE (checked 2026-11-10; details: https://agentgid.com/assets/liveportrait/)
```

Not verified: The license does not address generated videos directly, so YouTube use is left unset.

## Links
- AgentGid page: https://agentgid.com/assets/liveportrait/
- Is it free for commercial use? https://agentgid.com/assets/liveportrait/commercial-use/
- Source: https://huggingface.co/KlingTeam/LivePortrait
- Source: https://github.com/KlingAIResearch/LivePortrait
- Source: https://github.com/KlingAIResearch/LivePortrait/blob/main/LICENSE


## LivePortrait: make a portrait photo move like a video

LivePortrait is a free, open AI model that takes one portrait photo and a short video of someone moving their face. It makes the photo copy those expressions and head moves, so a still picture turns into a short clip. It runs on your own computer with a strong graphics card or an Apple Silicon Mac, and also works for cats and dogs.

### Where to use it
- Bring a still portrait to life: Take a photo or a painting of a face and a short clip of yourself making faces. You get a video where the portrait smiles, blinks and turns its head the same way.
- Change expressions in a finished video: Use a talking-head video instead of a photo and give it new facial expressions from another clip, without filming again.
- Make a pet photo wink or sing: Animate a photo of a cat or a dog with a person's face movements, for a fun social post or a greeting.
- Reuse motion without sharing a face: Save only the movement of a clip as a small file, then use it again and again. The file does not carry the face of the person who acted it.
- Edit a face by clicking in a browser: Open the local web interface to change eyes, lips or head pose of a portrait with controls instead of commands.
- A friendly presenter for course intros: Animate a drawn mascot or your own photo to greet viewers at the start of a lesson or product video. (idea, not from the official pages)

### How to start
- Get the code and make a Python 3.10 environment. Git, conda and FFmpeg must already be installed.

```
git clone https://github.com/KlingTeam/LivePortrait
cd LivePortrait

# create env using conda
conda create -n LivePortrait python=3.10
conda activate LivePortrait
```

- Install the packages for your machine; Apple Silicon Macs use their own list.

```
pip install -r requirements.txt
# for macOS with Apple Silicon users
pip install -r requirements_macOS.txt
```

- Download the model files from Hugging Face into pretrained_weights, then run the example.

```
huggingface-cli download KlingTeam/LivePortrait --local-dir pretrained_weights --exclude "*.git*" "README.md" "docs"
```


### License
The LivePortrait code and models are under the MIT license. The face detection models it ships with, from InsightFace, are for non-commercial research only, so swap them before any paid use.

### Good to know
- You need a computer with an NVIDIA graphics card, or an Apple Silicon Mac. On a Mac it may be about 20 times slower.
- The animal mode was only tested on Linux and Windows with an NVIDIA card. It does not work on a Mac.
- Results look best when your motion clip is square, shows mostly the head and starts with a calm face looking straight ahead.
- Moving a real person's face can be misused for fakes. The makers ask for responsible use and take no legal responsibility for results.
- There is a free online demo on Hugging Face if you want to try it before setting it up.

Checked 2026-11-10 against: https://github.com/KlingAIResearch/LivePortrait, https://github.com/KlingAIResearch/LivePortrait/blob/main/readme.md, https://github.com/KlingAIResearch/LivePortrait/blob/main/LICENSE, https://github.com/KlingAIResearch/LivePortrait/blob/main/assets/docs/changelog/2024-07-10.md, https://github.com/KlingAIResearch/LivePortrait/blob/main/assets/docs/changelog/2024-08-19.md, https://github.com/KlingAIResearch/LivePortrait/blob/main/assets/docs/speed.md, https://github.com/KlingAIResearch/LivePortrait/blob/main/requirements_base.txt, https://huggingface.co/KlingTeam/LivePortrait, https://huggingface.co/KlingTeam/LivePortrait/raw/main/README.md, https://liveportrait.github.io/

### Mistakes to avoid
- Don't: shipping the default pipeline in a paid app or client project → Do: replace the InsightFace detection models (buffalo_l) with a commercially usable face detector first (The InsightFace models are non-commercial research only, even though the LivePortrait code is MIT.)
- Don't: python inference.py on an Apple Silicon Mac without the fallback flag → Do: PYTORCH_ENABLE_MPS_FALLBACK=1 python inference.py (The README gives this environment variable for Apple Silicon runs.)
- Don't: using a wide, uncropped driving video with lots of shoulder movement → Do: crop it to 1:1 around the head, or add --flag_crop_driving_video (The authors recommend a square head crop with little shoulder movement.)
- Don't: accepting a result with a black box over the face → Do: run again with --no_flag_use_half_precision (Half precision causes the black box problem on some machines.)
- Don't: python app.py --flag_do_torch_compile on Windows or macOS → Do: use plain python app.py there; the compile speed-up is for Linux with NVIDIA (The README says this option is not supported on Windows and macOS.)

### Questions
- Can I use LivePortrait in a paid product? Only after you replace the InsightFace face detection models, which are for non-commercial research only. The rest is under the MIT license.
- Does it work on a Mac? Yes for people on Apple Silicon, but slower; the animal mode is not supported on a Mac.
- Can a photo drive another photo, not only a video? Yes. You can give an image as the motion source and limit the change to lips, eyes, pose or expression.
- Does the sound of my motion clip stay? Yes. If the driving video has sound, it is added to the result, at the same frame rate.

---
AgentGid's catalog is curated: every entry was checked against the maker's official pages (registry, repository, license and pricing pages) on the date shown; a license value shows the quote it stands on wherever AgentGid has one. Usage numbers come from AgentGid's own scan of public repositories set up for AI coding agents. Claims by makers are reported, not tested by AgentGid.
