DFACE.app (adam.harvey.studio)

DFACE.app is a lightweight, open-source web app for automatic, private face redaction that runs entirely in your browser so images never leave your device. Built as part of the VFRAME project, it uses a YOLOv5-based object detector ported to run with TF.js and Next.js to find and redact faces at scale—supporting batch processing, applying effects (color fill, blur, or emoji), and detecting up to 1,000 faces per image down to roughly 10×10 pixels. The single-page app is small (~13 MB of bandwidth), published under an MIT license, and the source is available on GitHub. For the AI/ML community the tool is significant because it demonstrates practical, client-side deployment of a modern object detector for privacy-preserving workflows. It showcases how inference-heavy models like YOLOv5 can be adapted to run in-browser with acceptable performance and resource footprints, enabling activist and social-media use cases where data sovereignty matters. DFACE also illustrates an accessible model-to-web pipeline (YOLOv5 → TF.js) and is a reusable reference implementation for researchers and developers looking to build privacy-first computer vision tooling; its development was supported by the NGI0 PET Fund.