Video Editing Toolkit

Goal

Complete video production pipeline: silence removal, auto-captions, vertical cropping, YouTube clipping, compression, and 3D transitions.

Scripts (7 total)

Quick Start Recipes

Recipe 1: Full Social Media Pipeline

# 1. Remove silences
python3 ./scripts/jump_cut_vad_singlepass.py input.mp4 .tmp/edited.mp4

# 2. Add captions
python3 ./scripts/auto_captions.py .tmp/edited.mp4 .tmp/captioned.mp4 --word-level --max-words 2

# 3. Crop to vertical for Shorts/Reels
python3 ./scripts/vertical_crop.py .tmp/captioned.mp4 .tmp/vertical.mp4

# 4. Compress for platform
python3 ./scripts/compress_video.py .tmp/vertical.mp4 output.mp4 --preset instagram-reel

Recipe 2: YouTube → Shorts Pipeline

# 1. Download + auto-clip YouTube video
python3 ./scripts/youtube_clip.py "https://youtube.com/watch?v=..." --auto-clip --output-dir clips/

# 2. Add captions to best clip
python3 ./scripts/auto_captions.py clips/chapters/01_intro.mp4 .tmp/captioned.mp4 --word-level

# 3. Crop to vertical
python3 ./scripts/vertical_crop.py .tmp/captioned.mp4 .tmp/vertical.mp4

# 4. Compress for YouTube Shorts
python3 ./scripts/compress_video.py .tmp/vertical.mp4 short.mp4 --preset youtube-shorts

Recipe 3: Quick Edit + Upload

# All-in-one: silence removal + transcription + metadata + Auphonic upload
python3 ./scripts/simple_video_edit.py --video input.mp4 --title "My Video"

Script 1: VAD Silence Removal

File: ./scripts/jump_cut_vad_singlepass.py

How It Works

1. Extracts audio as WAV (16kHz mono)
2. Runs Silero VAD to detect speech segments
3. Merges close segments, adds padding
4. Uses FFmpeg trim+concat to join segments in single pass
5. Hardware encodes with hevc_videotoolbox (H.265, 17Mbps, 30fps)

CLI Arguments

Usage

python3 ./scripts/jump_cut_vad_singlepass.py input.mp4 output.mp4
python3 ./scripts/jump_cut_vad_singlepass.py input.mp4 output.mp4 --min-silence 1.0 --padding 200

Processing Time

~8 minutes for a 49-min 4K video

Script 2: Auto-Captions

File: ./scripts/auto_captions.py

How It Works

1. Transcribes video using faster-whisper (word-level timestamps)
2. Generates SRT subtitles (segment-level or word-level)
3. Burns styled captions into video with FFmpeg

CLI Arguments

Usage

# Basic captions
python3 ./scripts/auto_captions.py input.mp4 output.mp4

# Word-level (CapCut/Submagic style)
python3 ./scripts/auto_captions.py input.mp4 output.mp4 --word-level --max-words 2

# Yellow captions at top
python3 ./scripts/auto_captions.py input.mp4 output.mp4 --font-color yellow --position top

# SRT only (no burn)
python3 ./scripts/auto_captions.py input.mp4 --srt-only

# High accuracy
python3 ./scripts/auto_captions.py input.mp4 output.mp4 --model large-v3

Dependencies

pip install faster-whisper

Script 3: Vertical Crop

File: ./scripts/vertical_crop.py

How It Works

1. Samples frames throughout the video
2. Detects faces using OpenCV Haar cascade
3. Smooths face positions to avoid jitter
4. Crops 16:9 → 9:16 centered on the speaker
5. For moving subjects: segments video into 2s chunks with per-segment tracking

CLI Arguments

Usage

# Auto face-tracking crop
python3 ./scripts/vertical_crop.py input.mp4 output.mp4

# Square crop (Instagram post)
python3 ./scripts/vertical_crop.py input.mp4 output.mp4 --ratio 1:1

# 4:5 crop (Instagram feed)
python3 ./scripts/vertical_crop.py input.mp4 output.mp4 --ratio 4:5

# Center crop (no tracking)
python3 ./scripts/vertical_crop.py input.mp4 output.mp4 --position center

Dependencies

pip install opencv-python numpy

Script 4: YouTube Clip

File: ./scripts/youtube_clip.py

How It Works

1. Downloads video via yt-dlp (up to 4K)
2. Extracts YouTube chapters if available
3. Falls back to AI chapter generation (Whisper + Claude)
4. Clips video into individual chapter files

CLI Arguments

Usage

# Download only
python3 ./scripts/youtube_clip.py "https://youtube.com/watch?v=..." --download-only

# Auto-clip into chapters
python3 ./scripts/youtube_clip.py "https://youtube.com/watch?v=..." --auto-clip

# Extract specific range
python3 ./scripts/youtube_clip.py "https://youtube.com/watch?v=..." --start 60 --end 180 -o clip.mp4

# Download audio only
python3 ./scripts/youtube_clip.py "https://youtube.com/watch?v=..." --audio-only

# Top 3 clips only
python3 ./scripts/youtube_clip.py "https://youtube.com/watch?v=..." --auto-clip --max-clips 3

Dependencies

pip install yt-dlp faster-whisper anthropic

Script 5: Video Compressor

File: ./scripts/compress_video.py

Platform Presets

CLI Arguments

Usage

# Platform preset
python3 ./scripts/compress_video.py input.mp4 output.mp4 --preset instagram-reel
python3 ./scripts/compress_video.py input.mp4 output.mp4 --preset whatsapp

# Target file size
python3 ./scripts/compress_video.py input.mp4 output.mp4 --target-size 25

# Custom
python3 ./scripts/compress_video.py input.mp4 output.mp4 --resolution 720 --crf 28

# List all presets
python3 ./scripts/compress_video.py input.mp4 output.mp4 --list-presets

Script 6: Simple Video Edit (Full Pipeline)

File: ./scripts/simple_video_edit.py

How It Works

1. FFmpeg silence detection + cutting
2. Audio normalization (loudnorm)
3. Whisper transcription
4. Claude-generated YouTube metadata (summary + chapters)
5. Auphonic upload → YouTube (private draft)

CLI Arguments

Usage

# Full pipeline
python3 ./scripts/simple_video_edit.py --video input.mp4 --title "My Video"

# Local only
python3 ./scripts/simple_video_edit.py --video input.mp4 --title "Test" --no-upload

Dependencies

pip install anthropic faster-whisper requests python-dotenv
# Requires: ANTHROPIC_API_KEY, AUPHONIC_API_KEY in .env

Script 7: 3D Swivel Teaser

File: ./scripts/insert_3d_transition.py

How It Works

1. Extracts frames from later in video (default: 60s onwards)
2. Creates 3D rotating "swivel" animation via Remotion
3. Splits video: intro → transition → main content
4. Re-encodes and concatenates with audio preserved

CLI Arguments

Final Timeline

[0-3s intro] [3-8s swivel teaser @ 100x] [8s onwards: edited content]
Audio: Original audio plays continuously

Dependencies

pip install torch  # For Silero VAD (jump_cut script)
brew install ffmpeg node  # macOS
cd video_effects && npm install  # For Remotion 3D rendering

All Dependencies (Install Once)

# Core (required for all scripts)
brew install ffmpeg

# Silence removal
pip install torch

# Auto-captions + YouTube clip
pip install faster-whisper

# Vertical crop
pip install opencv-python numpy

# YouTube download
pip install yt-dlp

# AI features (chapters, metadata)
pip install anthropic

# Full pipeline (simple_video_edit)
pip install requests python-dotenv

# 3D transitions
brew install node
cd scripts/video_effects && npm install

Troubleshooting

Technical Details

• macOS: Hardware encoding (hevc_videotoolbox / h264_videotoolbox)
• Fallback: libx264/libx265 with CRF
• Audio: AAC 128-192kbps
• Uses hvc1 codec tag for QuickTime compatibility
• All scripts support --help for full argument list

Schema

Inputs

Outputs

Credentials

Composable With

Skills that chain well with this one: pan-3d-transition, recreate-thumbnails

Cost

Free locally (FFmpeg + Silero VAD)