VHS Home Video Scene Detection With PySceneDetect

A typical VHS tape recorded over a decade of family life has hundreds of distinct moments buried in it — birthday parties, holidays, random Tuesday afternoons in the backyard. Finding a specific one means scrubbing through an hour of footage hoping you land in the right place.

Scene detection solves this. It automatically identifies where one scene ends and another begins — usually at camera cuts or significant changes in the frame — and splits your footage into individual clips. Instead of one 90-minute file, you get 200 clips, each a distinct moment.

What Is PySceneDetect?

PySceneDetect is a free, open-source Python tool that analyzes video frame-by-frame and detects scene boundaries. It works by looking for sudden changes in color, brightness, or content between frames — the kind of change that happens when someone stops filming and starts again.

For VHS footage this works surprisingly well, even accounting for tape noise and the slightly washed-out look of old video.

Installation

pip install scenedetect[opencv]

You'll also need ffmpeg installed. On Mac: brew install ffmpeg. On Windows, grab it from ffmpeg.org and add it to your PATH.

Basic Scene Detection

To detect scenes in a video file:

scenedetect -i "1990_summer.mp4" detect-content list-scenes

This outputs a CSV file listing the start time, end time, and duration of every detected scene. For a 90-minute tape you might get 150–300 scenes depending on how actively the camera was used.

Splitting Into Clips

To actually split the video into individual clip files:

scenedetect -i "1990_summer.mp4" detect-content split-video

This creates a folder of MP4 files, one per scene. Each file is named with its scene number and timestamp, like 1990_summer-Scene-001-00-00-00.mp4.

Tuning for VHS Footage

VHS tape noise can cause false positives — random frame glitches getting detected as scene changes. A few adjustments help:

scenedetect -i "1990_summer.mp4" detect-content \
  --threshold 35 --min-scene-len 3s split-video

• --threshold 35 — higher values require a bigger visual change to trigger a scene break (default is 27; 30–40 works well for VHS)
• --min-scene-len 3s — ignores any "scene" shorter than 3 seconds, filtering out noise glitches

What to Do With the Clips

Once you have individual scene clips, you can:

• Run Whisper on each one for per-scene transcription
• Extract a thumbnail frame for a visual gallery
• Name and organize clips manually by person or event
• Share specific clips directly without sending the full tape

Combining scene detection with transcription (see our Whisper guide) gives you a fully indexed archive — every moment findable by what was said or shown.

The Time Investment

Scene detection itself is fast — a 90-minute tape processes in a few minutes on most modern computers. The manual step is reviewing the output and naming things meaningfully. For a collection of 20–30 tapes, expect to spend a few evenings getting everything organized.