In the midst of building a climbing app, I want to use climbing videos of other climbers but I didn't want to show their face without their consent so I tried to look for quick solution with ffmpeg to blur human faces. There is https://github.com/ORB-HD/deface. But why use a ready made software when you can have AI build one for you

This project uses media pipe pose detection as a fallback of face detection. The app is hosted on github page as it is completely static. I've only tested in chrome and firefox, with climbing videos (medium range). The detection is jittery so manual editing is supported. The mask can be separately downloaded as a webp, and a pairing ffmpeg command is provided which can be run locally and much faster.

Source code at https://github.com/kmcheung12/faceblind