Ward – reliable face unlock for Linux login, sudo and polkit
Ward, a new open-source application, enables secure face unlock for Linux systems using infrared and RGB cameras with liveness detection.
Ward, a new open-source application, brings secure face unlock functionality to Linux systems. The application uses both infrared and RGB cameras to authenticate users, backed by real liveness checks, and works natively on Wayland. Unlike previous solutions, Ward always falls back to a password if face recognition fails, ensuring users never get locked out of their systems.
Solving Howdy's Issues
While Howdy brought Windows Hello-style face login to Linux, it had reliability issues unrelated to face recognition technology. Ward addresses these problems by shipping its own engine based on proven components like dlib, OpenCV, and V4L2. The new application properly finds and activates the infrared emitter, enrolls multiple poses during setup, and sets recognition thresholds based on the user's actual face measurements.
Key Features
Ward works effectively in complete darkness by utilizing the infrared camera and emitter. It includes real liveness detection that prevents replayed faces from being authenticated. The application measures recognition thresholds rather than guessing them, guided by enrollment that captures multiple poses. Ward offers both a native GTK 4/libadwaita interface and a command-line interface, with the latter supporting headless operation over SSH.
Installation and Compatibility
Ward can be installed directly from distribution repositories for Ubuntu, Fedora, and others. It uses the distribution's native Python installation without requiring pip dependencies. The application has been tested on Debian, Ubuntu, Fedora, Arch, and openSUSE. While it works best on laptops with Windows Hello infrared cameras (2019+ business/premium models), it also functions with standard webcams, though without infrared liveness detection in that case.
Security Considerations
As a 2D face recognition system, Ward has specific limitations. It defeats phone-screen photos and works in darkness but cannot defeat high-quality printed infrared photographs. Identical twins can bypass any 2D system. The developers emphasize that Ward should be treated as strong convenience with a password fallback, similar to Windows Hello on the same hardware, rather than a security system like Face ID. No false-accept rate has been certified yet.