Face-Tracking Video Reframe

Crop a video to a target aspect ratio so it matches other clips (e.g. a horizontal 848x384 clip into a 9:16 portrait), while a dynamic crop window tracks the speaker's face — the face stays centered and is never cut off, even as the person moves.

This solves the problem a static center-crop can't: when the face drifts left/right (or up/down) over the clip, a fixed window clips it at the edges. Here the window follows the detected face on a smoothed path.

Requirements

• ffmpeg + ffprobe on PATH
• Python with opencv-python and numpy
  • check: python -c "import cv2, numpy" — if it fails: pip install opencv-python numpy

How to Use

The script does everything (detect → smooth track → crop → scale → mux original audio).