- Home
- AI Talking Photo Apps
- SadTalker
SadTalker
What it is
SadTalker creates talking-head videos from a single portrait image and speech audio. It derives 3D head-pose and expression coefficients from the audio to animate a face, with still, reference, and resize modes as well as full-image animation. Command-line examples include GFPGAN enhancement. The project links to Hugging Face and Colab demos and documents local use through a Gradio demo or command-line inference on Windows, Linux, and macOS. Setup has dependencies: Linux instructions specify Python 3.8, PyTorch, FFmpeg, and repository requirements; Windows instructions also call for Python, Git, FFmpeg, and model files. Pre-trained models are linked from Google Drive, GitHub Releases, and Baidu. The project says its open-source code can run offline and does not collect personal information or other data. SadTalker is free under Apache 2.0, except for listed third-party components, and its output is 720p or lower without a watermark. The README notes a Discord integration and a Stable Diffusion WebUI extension, and prohibits unlawful or harmful uses, including fraud and portrait-rights infringement.
For and against
For
- Free under Apache 2.0, with listed third-party exceptions.
- Code can run offline without collecting personal information.
- Includes still, reference, and resize modes.
- Local Gradio and command-line use are documented.
- Generated output has no watermark.
Against
- Output resolution is 720p or lower.
- Local setup requires dependencies and model files.
- Use restrictions prohibit unlawful or harmful purposes.
The verdict
SadTalker offers several ways to animate a portrait from speech, with local setup paths and online demos. Its output is limited to 720p or lower, and the project sets restrictions on use.
Who it suits
SadTalker suits people who want to generate talking-head videos from portrait images and speech audio, either locally or through linked online demos. Local setup is documented for Windows, Linux, and macOS.
Plans and prices
SadTalker open-source software
Apache 2.0 except listed third-party components · offline code
Read 4 Oct 2026 · github.com
On the record
- Enhancement
- The CLI examples support GFPGAN enhancement for generated videos.github.com · read 4 Oct 2026
- Integrations
- The README says SadTalker is integrated into Discord and offers a Stable Diffusion WebUI extension.github.com · read 4 Oct 2026
- License
- The repository license identifies Apache 2.0 as the SadTalker license, except for listed third-party components.github.com · read 4 Oct 2026
- Local use
- The README documents a local Gradio demo, command-line inference, and Windows, Linux, and macOS setup paths.github.com · read 4 Oct 2026
- Models
- Pre-trained models are linked for download from Google Drive, GitHub Releases, and Baidu.github.com · read 4 Oct 2026
- Modes
- The project describes still, reference, and resize modes, as well as full-image animation.github.com · read 4 Oct 2026
- Motion generation
- It generates 3D head-pose and expression coefficients from audio to animate a face.sadtalker.github.io · read 4 Oct 2026
- Offline operation
- The project says the open-source code runs completely offline and collects no personal information or other data.github.com · read 4 Oct 2026
- Online demos
- The project links to Hugging Face and Colab demos.sadtalker.github.io · read 4 Oct 2026
- Purpose
- SadTalker generates a talking-head video from a single portrait image and speech audio.github.com · read 4 Oct 2026
- Research context
- The project page identifies SadTalker as a CVPR 2023 work by researchers affiliated with Xi'an Jiaotong University, Tencent AI Lab, and Ant Group.sadtalker.github.io · read 4 Oct 2026
- Setup requirements
- The documented Linux setup uses Python 3.8, PyTorch, FFmpeg, and the repository requirements; Windows setup also requires Python, Git, FFmpeg, and model files.github.com · read 4 Oct 2026
- Support
- The README directs users to its FAQs before opening an issue and links to project discussions.github.com · read 4 Oct 2026
- Use restrictions
- The README prohibits use that harms others' rights or violates applicable law or social ethics, including fraud and portrait-rights infringement.github.com · read 4 Oct 2026
