Video Editing Toolkit

Goal

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

Scripts (7 total)

Script	Purpose
jump_cut_vad_singlepass.py	Remove silences with neural VAD (Silero)
auto_captions.py	Generate + burn styled subtitles (Whisper + FFmpeg)
vertical_crop.py	Auto-crop 16:9 → 9:16 with face tracking
youtube_clip.py	Download YouTube + AI chapter clipping
compress_video.py	Social media compression with platform presets
simple_video_edit.py	Full pipeline: silence removal + transcription + metadata + upload
insert_3d_transition.py	3D swivel teaser insertion

---

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

Argument	Default	Description
--min-silence	0.5	Min silence duration to cut (seconds)
--min-speech	0.25	Min speech duration to keep (seconds)
--padding	100	Padding around speech (ms)
--merge-gap	0.3	Merge segments closer than this (seconds)
--keep-start	true	Always start from 0:00

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

Argument	Default	Description
--model	base	Whisper model: tiny, base, small, medium, large-v3
--language	auto	Force language (en, es, hi, etc.)
--word-level	false	Short punchy captions (1-3 words at a time)
--max-words	3	Max words per caption in word-level mode
--font-size	22	Caption font size (auto-scales for vertical)
--font-color	white	Text color: white, yellow, cyan, green, red, orange, pink
--outline-color	black	Outline color
--position	bottom	Caption position: bottom, top, middle
--srt-only	false	Only generate SRT, don't burn
--srt-path	auto	Custom SRT output path
--bold	true	Bold text

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

Argument	Default	Description
--ratio	9:16	Target aspect ratio (e.g., 9:16, 4:5, 1:1)
--position	auto	Crop position: auto (face tracking), left, center, right
--smoothing	30	Smoothing window in frames
--sample-interval	5	Sample every N frames for detection

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

Argument	Default	Description
--download-only	false	Only download, don't clip
--max-quality	1080	Max quality: 480, 720, 1080, 1440, 2160
--audio-only	false	Download audio only (MP3)
--start	none	Manual clip start (seconds)
--end	none	Manual clip end (seconds)
--auto-clip	false	AI-powered chapter detection and clipping
--use-yt-chapters	true	Use YouTube chapters if available
--max-clips	none	Limit number of clips
--whisper-model	base	Whisper model for transcription
--reencode	false	Re-encode clips (precise cuts)
--output-dir	.tmp/clips	Output directory

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

Preset	Resolution	CRF	Max Duration	Max Size	Notes
youtube	1080p	18	none	none	High quality
youtube-shorts	1080x1920	20	60s	none	Vertical
instagram-reel	1080x1920	23	90s	none	Vertical
instagram-post	1080x1080	23	60s	none	Square/vertical
tiktok	1080x1920	23	180s	none	Vertical
twitter	1080p	23	140s	512MB	Auto-bitrate
linkedin	1080p	23	600s	200MB	Auto-bitrate
telegram	720p	28	none	50MB	For bots
whatsapp	720p	28	120s	16MB	Aggressive
small	480p	30	none	none	Quick sharing

CLI Arguments

Argument	Default	Description
--preset	none	Platform preset (see table above)
--resolution	1080	Max height in pixels
--crf	23	Quality (0=lossless, 51=worst)
--audio-bitrate	128k	Audio bitrate
--target-size	none	Target file size in MB
--max-duration	none	Max duration in seconds
--list-presets	false	Show all presets

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

Argument	Default	Description
--video	required	Input video path
--title	required	YouTube title
--thumbnail	none	Thumbnail image path
--no-upload	false	Skip Auphonic upload
--no-normalize	false	Skip audio normalization
--upload-only	false	Skip editing, just upload
--silence-threshold	-35	Silence threshold in dB
--silence-duration	3.0	Min silence duration (seconds)

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

Argument	Default	Description
--insert-at	3	Where to insert teaser (seconds)
--duration	5	Teaser duration (seconds)
--teaser-start	60	Where to sample content from (seconds)
--bg-color	#2d3436	Background color (hex)
--bg-image	none	Background image path

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

Issue	Solution
Cuts feel abrupt	--padding 200 in jump_cut
Too much cut	--min-silence 1.0 in jump_cut
Captions too small	--font-size 32 in auto_captions
Vertical video detected	Font auto-scales 1.3x
Won't play in QuickTime	Ensure hvc1 codec tag
Face not detected	Try --position center in vertical_crop
YouTube download fails	Update yt-dlp: pip install -U yt-dlp
File too large	Use --preset whatsapp or --target-size 25
Hardware encoder fails	Auto-falls back to software (libx264)

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

Name	Type	Required	Description
input_video	file_path	Yes	Input video file path
recipe	string	No	Pipeline recipe: full-social, youtube-shorts, quick-edit
preset	string	No	Compression preset: youtube, instagram-reel, tiktok, whatsapp, etc.

Outputs

Name	Type	Description
output_video	file_path	Processed video file path

Credentials

Name	Source
ANTHROPIC_API_KEY	.env (for AI chapters)
AUPHONIC_API_KEY	.env (for upload)

Composable With

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

Cost

Free locally (FFmpeg + Silero VAD)