akrindev/highlight-scanner

Highlight Scanner

This skill combines all detection methods to find viral-worthy highlights from videos. It's the core analysis component for the autocut-shorts workflow.

When to Use

• User wants to find the best moments from a video
• Identifying viral-worthy segments for short-form content
• Creating highlight reels from long videos
• Analyzing podcast, vlog, gaming, or tutorial content
• Preparing content for autocut workflow

Detection Signals

1. Transcript Analysis

• Identifies hooks and attention-grabbing phrases
• Detects story beats and important points
• Finds question/answer patterns
• Keyword matching for viral phrases

2. Laughter Detection

• Finds humorous moments
• Detects audience reactions
• Identifies funny segments

3. Sentiment/Emotion Analysis

• Positive emotions (excitement, joy)
• Surprise moments
• Negative emotions (controversy, drama)
• Emotional peaks and intensity

4. Scene Detection

• Scene changes as natural cut points
• Topic transitions
• Visual changes

Scoring System

Each highlight is scored based on:

virality_score = (
    transcript_score * 0.35 +
    laughter_score * 0.25 +
    sentiment_score * 0.25 +
    scene_score * 0.15
)

Score Range: 0.0 - 1.0

• 0.8 - 1.0: Premium viral potential (must use)
• 0.6 - 0.8: High potential (excellent clips)
• 0.4 - 0.6: Good potential (consider using)
• 0.2 - 0.4: Moderate potential (optional)
• 0.0 - 0.2: Low potential (skip)

Available Scripts

scripts/find_highlights.py

Find viral-worthy highlight segments.

Usage:

python skills/highlight-scanner/scripts/find_highlights.py <video_path> [options]

Options:

• --transcript-path: Path to transcript SRT/VTT file
• --scenes-path: Path to scenes JSON file (from scene-detector)
• --laughter-path: Path to laughter JSON file (from laughter-detector)
• --sentiment-path: Path to sentiment JSON file (from sentiment-analyzer)
• --num-clips: Number of clips to generate - default: 5
• --min-duration: Minimum clip duration (seconds) - default: 15
• --max-duration: Maximum clip duration (seconds) - default: 60
• --output, -o: Output JSON path (default: <video_path>_highlights.json)

Examples:

Find highlights with transcript only:

python skills/highlight-scanner/scripts/find_highlights.py video.mp4 --transcript-path video.srt

Full analysis with all signals:

python skills/highlight-scanner/scripts/find_highlights.py video.mp4 \
  --transcript-path video.srt \
  --scenes-path video_scenes.json \
  --laughter-path video_laughter.json \
  --sentiment-path video_sentiment.json

Find 10 clips with custom duration:

python skills/highlight-scanner/scripts/find_highlights.py video.mp4 \
  --transcript-path video.srt \
  --num-clips 10 \
  --min-duration 20 \
  --max-duration 45

scripts/analyze_viral_potential.py

Analyze the viral potential of video segments.

Usage:

python skills/highlight-scanner/scripts/analyze_viral_potential.py <video_path> [options]

Options:

• --transcript-path: Path to transcript file
• --output, -o: Output JSON path

Example:

python skills/highlight-scanner/scripts/analyze_viral_potential.py video.mp4 --transcript-path video.srt

Output Format

{
  "video_path": "video.mp4",
  "total_segments_analyzed": 15,
  "highlights": [
    {
      "rank": 1,
      "start_time": 45.2,
      "end_time": 72.5,
      "duration": 27.3,
      "virality_score": 0.92,
      "scores": {
        "transcript": 0.95,
        "laughter": 0.80,
        "sentiment": 0.85,
        "scenes": 0.70
      },
      "text": "This is the key moment text...",
      "reasoning": "Contains hook + laughter + positive emotion",
      "suggested_clip_start": 42.0,
      "suggested_clip_end": 75.0,
      "confidence": "high"
    }
  ],
  "analysis_summary": {
    "total_duration": 120.5,
    "avg_virality_score": 0.68,
    "best_segment_start": 45.2,
    "recommended_num_clips": 5
  }
}

Scoring Weights

Default weights (customizable):

DEFAULT_WEIGHTS = {
    'transcript': 0.35,   # Content analysis
    'laughter': 0.25,     # Humor detection
    'sentiment': 0.25,    # Emotion analysis
    'scenes': 0.15        # Visual transitions
}

Adjust weights based on content type:

• Comedy content: Increase laughter weight
• Emotional content: Increase sentiment weight
• Educational content: Increase transcript weight
• Action content: Increase scenes weight

Viral Phrases/Keywords

High-Viral Potential Phrases

Hooks/Attention Grabbers:

• "You won't believe..."
• "This changes everything..."
• "The secret to..."
• "What nobody tells you about..."
• "I made a huge mistake..."
• "This is illegal..."

Story Beats:

• "The plot twist..."
• "And then it happened..."
• "But here's the catch..."
• "The most important part..."

Engagement:

• "Comment if you agree..."
• "Like if you've experienced this..."
• "Wait for it..."
• "Watch till the end..."

Moderate-Viral Potential

• Interesting facts
• Tips and tricks
• How-to content
• Before/after reveals

Integration with Other Skills

This skill combines inputs from:

• video-transcriber: Transcript for content analysis
• scene-detector: Scene changes for cut points
• laughter-detector: Humorous moments
• sentiment-analyzer: Emotional peaks

Output is used by:

• video-trimmer: Create clips from highlights
• autocut-shorts: Full workflow execution

Common Workflow

1. User provides video file
2. Transcribe with video-transcriber
3. Detect scenes with scene-detector (optional)
4. Detect laughter with laughter-detector (optional)
5. Analyze sentiment with sentiment-analyzer (optional)
6. Find highlights using this skill (combines all signals)
7. Create clips from highlights with video-trimmer or autocut-shorts

Tips

• More input signals = better highlight detection
• Always provide transcript (minimum requirement)
• Scene detection helps with clean cuts
• Laughter detection improves viral potential
• Sentiment analysis identifies emotional peaks
• Adjust weights based on your content type
• Filter by score threshold for quality control
• Consider clip duration when selecting highlights

Performance

• Transcript only: ~2 seconds for 1-minute video
• Full analysis: ~10-30 seconds for 10-minute video
• Scales linearly with video duration
• Can process in real-time for live content

References

• Viral content analysis research
• Engagement metrics studies
• TikTok/YouTube algorithm insights