scene-detector

name	scene-detector
description	Detect scene changes and shot boundaries in videos. Use when you need to identify where scenes change, find natural cut points, or segment video into scenes. Supports adaptive detection for both fast cuts and gradual fades.
allowed-tools	Bash(ffmpeg:*) Bash(python:*)
compatibility	Requires PySceneDetect, FFmpeg, and OpenCV
metadata	{"version":"1.0","method":"Adaptive Content-Aware Detection"}

Scene Detector

This skill enables AI agents to detect scene changes and shot boundaries in videos using PySceneDetect.

When to Use

• User wants to find scene changes in a video
• Need natural cut points for video editing
• Want to segment long video into scenes
• Detecting transitions for autocut workflow

Available Scripts

scripts/detect_scenes.py

Detect scene changes in video.

Usage:

python skills/scene-detector/scripts/detect_scenes.py <video_path> [options]

Options:

• --threshold: Detection threshold (0.1-0.5) - default: 0.3 (lower = more scenes)
• --min-scene-len: Minimum scene length in seconds - default: 0.5
• --output, -o: Output JSON path (default: <video_path>_scenes.json)
• --split: Split video into clips
• --output-dir: Directory for split clips (default: ./clips/)

Examples:

Detect scenes with default settings:

python skills/scene-detector/scripts/detect_scenes.py video.mp4

Detect with more sensitivity (more scenes):

python skills/scene-detector/scripts/detect_scenes.py video.mp4 --threshold 0.2

Detect and split video into clips:

python skills/scene-detector/scripts/detect_scenes.py video.mp4 --split --output-dir ./scene_clips/

Detection Methods

Adaptive Detection (Default)

Adapts to video content, works well for:

• Fast cuts (music videos, action)
• Gradual fades (narrative content)
• Mixed content

Threshold-Based Detection

Fixed threshold comparison:

• --threshold 0.1: Very sensitive (many scenes)
• --threshold 0.3: Balanced (default)
• --threshold 0.5: Less sensitive (fewer scenes)

Output Format

JSON Output

{
  "video_path": "video.mp4",
  "duration": 123.45,
  "total_scenes": 15,
  "scenes": [
    {
      "scene_number": 1,
      "start_time": 0.0,
      "end_time": 8.5,
      "duration": 8.5,
      "content_type": "fast_cut"
    },
    {
      "scene_number": 2,
      "start_time": 8.5,
      "end_time": 23.2,
      "duration": 14.7,
      "content_type": "fade"
    }
  ]
}

Split Clips

When --split is used, creates:

clips/
  scene_001.mp4
  scene_002.mp4
  scene_003.mp4
  ...

Integration with Other Skills

After scene detection, you can use these skills:

• video-trimmer: Trim specific scenes
• highlight-scanner: Use scene changes as cut points
• autocut-shorts: Full workflow for creating short clips

Common Workflow

1. User provides video file
2. Detect scenes using this skill
3. Identify interesting scenes based on timing
4. Create short clips from selected scenes

Tips

• Lower threshold = more scenes detected
• Use --min-scene-len to filter out very short scenes
• Scene boundaries are excellent cut points for short-form content
• Consider scene changes when creating clips (15-60s each)

References

• PySceneDetect documentation: https://www.scenedetect.com/
• OpenCV documentation: https://docs.opencv.org/