video-add-chapters Checked

**4. Clean Transcript**

python clean_transcript.py "./output/merged_document.md" --backup

• Removes filler words
• Improves sentence structure
• Enhances paragraph cohesion
• Preserves timestamps and chapter boundaries

### Optional: Manual Review

If chapter boundaries need adjustment:

1. Edit chapters.json with corrected timestamps

2. Re-run generate_docs.py to regenerate documents

---

Highlight Video Generation (Optional)

After completing the chapter workflow, you can create a highlight video by selecting specific segments from the transcript.

### Method 1: Annotation in Transcript (Recommended)

Mark highlights directly in the transcript using <u> or == annotations.

flowchart LR
    A[Transcript] --> B["Add <u> or ==marks=="]
    B --> C[parse_highlight_annotations.py]
    C --> D[Highlight Script]
    D --> E[generate_highlights.py]
    E --> F[Highlight.mp4]

**Quick Start:**

# 1. Open transcript markdown and add annotations:
#    <u>text to highlight</u>  or  ==text to highlight==

# 2. Parse annotations to generate highlight script
python parse_highlight_annotations.py "transcript.md" --video "video.mp4"

# 3. (Optional) Edit the generated highlight script to add titles

# 4. Generate highlight video
python generate_highlights.py "transcript - highlight_script.md" --padding 0.5

**Supported Annotation Formats:**

• <u>highlighted text</u> - HTML underline (visible in most editors)
• ==highlighted text== - Markdown highlight (Obsidian compatible)

**Features:**

• Consecutive highlighted segments are automatically merged
• End times calculated from next segment start
• Auto-generates titles from first few words

### Method 2: Manual Script Editing

Export full transcript as editable script, then delete unwanted lines.

flowchart LR
    A[Transcript] --> B[Export Script]
    B --> C[User Deletes Lines]
    C --> D[Generate Video]
    D --> E[Highlight.mp4]

**Quick Start:**

# 1. Export editable highlight script
python export_highlight_script.py "video.mp4" \
  --transcript "./output/video - transcript.json"

# 2. Edit the script - delete unwanted lines (in any text editor)

# 3. Generate highlight video
python generate_highlights.py "./output/video - highlight_script.md"

### Generate Highlight Video

python generate_highlights.py "highlight_script.md" --output "highlights.mp4" --padding 0.5 --title-duration 3

• Parses [START-END] timestamps from script
• Adds 0.5s padding before/after each segment to avoid mid-sentence cuts
• Displays optional segment titles (yellow centered text, Korean font) for 3 seconds
• Merges all segments into single video using FFmpeg
• Output: {video} - highlights.mp4

### Highlight Script Format

# Highlight Script: Video Title

**Source Video**: /path/to/video.mp4

---

[00:00:09-00:00:21] {Gemini CLI 설정} 우선은 Gemini에다가 제가 현재 커뮤니티 볼트가...

[01:46-02:15] {설치 완료} 네 설치가 된 것 같습니다. 볼트에서 확인해 볼까요.

[03:56-04:01] 아웃박스 커뮤니티 폴더에 질문을 생성을 했어요.

**Format**: [START-END] {Optional Title} Text content

• Titles in {curly braces} are optional
• If provided, title appears as yellow centered text overlay (144px, Korean font supported)

---

File Structure

video-add-chapters/
├── SKILL.md                    # This document
├── requirements.txt            # Python dependencies
├── transcribe_video.py         # Step 1: Video → Transcript
├── suggest_chapters.py         # Step 2: Chapter boundary detection
├── generate_docs.py            # Step 3: Document generation
├── clean_transcript.py         # Step 4: Transcript cleaning
├── parse_highlight_annotations.py  # Parse <u> and == annotations from transcript
├── export_highlight_script.py      # Export transcript as editable highlight script
├── generate_highlights.py          # Generate highlight video from script
├── templates/                  # Markdown templates
│   ├── chapter.md
│   ├── index.md
│   └── youtube_chapters.txt
└── examples/                   # Sample outputs
    ├── sample_chapter.md
    └── sample_youtube_chapters.txt

Troubleshooting

| Issue | Cause | Solution |

|-------|-------|----------|

| Chapter boundaries don't match content | Boundary set at keyword first mention | Use transition signal patterns for boundaries |

| Merged document content mismatch | Manual updates missed in separate files | Update all related files when changing boundaries |

| Transcript timing seems off | Misdiagnosed as offset issue | Verify: Whisper timestamps = video timestamps (no offset) |

| Chapter content overlap | Boundary doesn't match content transition | Use end signals for endpoints, start signals for start points |

### Verification Checklist

1. [ ] Verify video timestamp at each chapter start

2. [ ] Confirm next chapter content doesn't start before current chapter ends

3. [ ] Check merged document body matches chapter boundaries

4. [ ] Test YouTube link timestamps are accurate

5. [ ] Verify original meaning is preserved after cleaning

Language Support

Currently optimized for **Korean** language with hardcoded transition patterns. Multi-language support: **TBA**.

Cost Estimate

Whisper API pricing: ~$0.006 per minute of audio

• 1-hour video: ~$0.36
• Processing is done in 15-minute chun