Clipper

Create viral vertical clips from YouTube videos with auto-cropping and captions.

Pipeline

1. Fetch transcript from YouTube (saved as full_transcript.txt + full_transcript.srt)
2. Use LLM to identify viral segments (outputs segments.json)
3. Download only the needed segments from YouTube (not the full video)
4. Auto-crop using MediaPipe face detection + add captions
5. Output clips to outputs/

Installation

pip install -r requirements.txt

Usage

python main.py "https://www.youtube.com/watch?v=VIDEO_ID"

Options

python main.py "URL" --segments path/to/segments.json  # custom segments file
python main.py "URL" --aspect vertical                  # vertical (default), square, horizontal
python main.py "URL" --captions bold                    # bold (default), minimal, colorful, subtle, none

Segments File

Clip timestamps are defined in segments.json. Each clip has:

{
  "timestamp_start": "00:01:37,359",
  "timestamp_end": "00:01:46,078",
  "suggested_title": "Clip Title",
  "viral_score": 9.0
}

Clips are processed in order of highest viral score first.

Output

outputs/
  full_transcript.txt        # full video transcript
  full_transcript.srt        # full video SRT
  clip_1_Clip_Title.mp4      # video clip
  clip_1_Clip_Title.srt      # clip SRT
  ...

Configuration

Edit config.py to customize:

• Aspect ratios: vertical (9:16), square (1:1), horizontal (16:9)
• Caption presets: minimal, bold, colorful, subtle
• FFmpeg settings: encoding preset, quality (CRF), audio bitrate

Caption Presets

Preset	Description
bold	Large text, lower half, single line
minimal	Small text, bottom of screen
colorful	Yellow text, centered
subtle	Small with background