Deep-Live-Cam
github.com/hacksider/deep-live-cam- Category
- AI Tools
- Rank
- No. 157Tools index
- Pricing
- Open Source
- Type
- APP
- Use case
- Design & Media
- Builder
- hacksider
- GitHub
- 96.8k stars
- Latest release
- 2.7.5-Ultimate
- Date
About
Real-time face swap and deepfake tool that uses AI to replace faces in live video streams or recordings using just a single source image. Includes built-in safety checks and supports multiple execution providers for different hardware configurations.
What it does
Deep-Live-Cam is a desktop media workstation for changing faces in camera feeds, images, and videos. Its controls cover face selection, camera choice, previews, output encoding, mouth preservation, enhancement, and assigning different faces to multiple people.
Why it's ranked here
The strongest case is practical breadth: one interface handles live cameras and saved media, while face mapping, mouth masking, enhancement, audio retention, and several hardware backends offer meaningful control. The setup burden and unmaintained command-line mode keep it from being an easy general-purpose component.
What's good
The live workflow is direct: choose a face, choose a camera, and start the preview. Face mapping handles multiple subjects, while the mouth mask can retain original mouth movement. Video processing can preserve frame rate, audio, or temporary frames and offers three output encoders.
Tradeoffs
Manual installation requires Python, pip, Git, FFmpeg, external model downloads, and platform-specific runtime work. GPU setup differs across CUDA, CoreML, DirectML, and OpenVINO, with strict dependency pairings in some cases. The command-line interface is explicitly unmaintained. Webcam streaming also requires a separate screen-capture tool such as OBS.
How to use it well
Use it for consented character performances, live shows, memes, or local experiments where an operator can manage models and hardware dependencies. Start with the graphical interface and validate the preview before recording or streaming. It does not provide its own broadcasting pipeline, and its command-line path is a poor foundation for maintained automation.
Technical notes+
run.py adjusts PATH and native-library discovery before delegating to modules/core.py; Windows receives CUDA and OpenVINO DLL handling, while Linux preloads NVIDIA shared objects. modules/core.py exposes argparse-driven headless processing, provider selection, memory limits, frame processors, encoding controls, and optional NSFW filtering. modules/ui.py implements the PySide6 GUI, persisted switches, camera enumeration, previews, and thread-safe status signals. requirements.txt pins OpenCV, ONNX, InsightFace, PySide6, OpenNSFW2, and platform-specific ONNX Runtime builds. benchmark_pipeline.py is a standalone webcam pipeline benchmark rather than a test suite.
Observed
- Primary language
- Python
- Interfaces
- PySide6 desktop GUI and an argparse command-line interface described as unmaintained
- Installation surface
- Manual virtual-environment installation from requirements.txt plus external ONNX model downloads; pre-built quickstart builds are also advertised
- Platform support
- Windows, Linux, and macOS instructions are provided, including Apple Silicon setup
- Execution backends
- CPU, CUDA, CoreML, DirectML, and OpenVINO configurations are documented
- Media dependencies
- FFmpeg is required; OpenCV, ONNX Runtime, InsightFace, and PySide6 are declared dependencies
- Structural observation
- A standalone 200-frame webcam pipeline benchmark is included in benchmark_pipeline.py
Read from README.md, pyproject.toml, requirements.txt, run.py, tkinter_fix.py, benchmark_pipeline.py, modules/ui.py, modules/run.py, modules/core.py, modules/paths.py, modules/typing.py, modules/gettext.py, modules/globals.py, modules/__init__.py, modules/capturer.py.
What it can do
Perform real-time face swap in live video streams
Live video feed and source face image → Live video stream with swapped face
Replace faces in pre-recorded video files
Video file and source face image → Video file with swapped face
Generate deepfake content using single source image
Single face photograph and target video/stream → Deepfake video or live stream
Execute face swapping on different hardware configurations
Face swap parameters and hardware provider selection → Optimized face swap processing
Apply built-in safety checks to prevent misuse
Source images and target content → Safety validation results and content restrictions
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.