LivePortrait
KuaishouLivePortrait is a portrait-animation model from Kuaishou's Kling research team that turns a single still image of a face, human or animal, into video driven by the expressions and head pose of a reference clip. It uses implicit keypoints with stitching and retargeting control rather than diffusion, which makes it fast enough for real-time use, and supports video-to-video editing and regional control.
Openness
2 high confidence- weights
- open(KlingTeam/LivePortrait downloads from Hugging Face without a gate, Humans and Animals models together)
- data
- described(the paper describes about 69 million training frames
- code
- partial(inference code only
- license
- MIT(the LivePortrait code and weights)+InsightFace-Non-Commercial-Research(the InsightFace face-detection models bundled in the weight tree are "for non-commercial research purposes only"
LivePortrait's own code and weights are MIT, but every run crops the face with InsightFace detection models licensed for non-commercial research only, and the one download command fetches them with the weights. The LICENSE says commercial users must swap them out. Only inference code is published and the training data is not released.
- https://arxiv.org/abs/2407.03168 recorded 2026-09-26
Paper abstract: "we scale up the training data to about 69 million high-quality frames, adopt a mixed image-video training strategy"; no dataset release.
- https://huggingface.co/api/models/KlingTeam/LivePortrait?expand[]=downloads&expand[]=likes&expand[]=cardData&expand[]=gated&expand[]=lastModified&expand[]=createdAt recorded 2026-09-26
Hugging Face API record: cardData license mit, gated false.
- https://raw.githubusercontent.com/KlingAIResearch/LivePortrait/main/readme.md recorded 2026-09-27
The download step is "huggingface-cli download KlingTeam/LivePortrait --local-dir pretrained_weights --exclude "*.git*" "README.md" "docs"", which fetches the whole weight tree, insightface/models/buffalo_l included.
- https://raw.githubusercontent.com/KlingAIResearch/LivePortrait/main/src/utils/cropper.py recorded 2026-09-27
The Cropper's constructor loads FaceAnalysisDIY(name="buffalo_l", root=self.crop_cfg.insightface_root), prepares it and warms it up; the face crop runs on these InsightFace models.
- https://ungh.cc/repos/KlingAIResearch/LivePortrait/files/main recorded 2026-09-26
Repository file list: app.py, inference.py and the model sources; InsightFace and X-Pose vendored as dependencies; no training script.
- https://ungh.cc/repos/KlingAIResearch/LivePortrait/files/main/LICENSE recorded 2026-09-26
LICENSE: "MIT License / Copyright (c) 2024 Kuaishou Visual Generation and Interaction Center", followed by "The models of InsightFace are for non-commercial research purposes only."
Adoption
2 high confidenceAdoption is measured as Hugging Face downloads of the LivePortrait checkpoint repository. The weights are also mirrored on Google Drive and Baidu, which this figure does not count.
- https://huggingface.co/api/models/KlingTeam/LivePortrait?expand[]=downloads&expand[]=likes&expand[]=cardData&expand[]=gated&expand[]=lastModified&expand[]=createdAt recorded 2026-09-26
Hugging Face API record for KlingTeam/LivePortrait: downloads 10340.
Capability
1 medium confidenceLivePortrait animates a still portrait from a driving video rather than generating video from a prompt, and no public arena ranks that task. It sits below Mochi, which at least places on the text-to-video arena.
- https://artificialanalysis.ai/video/leaderboard/text-to-video recorded 2026-09-26
The text-to-video leaderboard lists no LivePortrait entry.
- https://ungh.cc/repos/KlingAIResearch/LivePortrait/files/main/readme.md recorded 2026-09-26
README: "the official PyTorch implementation of our paper LivePortrait: Efficient Portrait Animation with Stitching and Retargeting Control", with image-driven and video-to-video modes.
Verified 2026-09-26