YouTube Shorts Creator

A powerful tool to automatically convert long-form videos into engaging YouTube Shorts using AI analysis and professional video effects.

Detailed description about this project can be found in the blog post. 👨‍💻 Want to learn more about the author? Check out vitaliihonchar.com or the YouTube channel for more content!

Features

• 🎯 AI-Powered Content Analysis - Uses OpenAI/OpenRouter to identify the best moments for shorts with strict duration requirements
• 🎬 Professional Video Effects Pipeline - Automated video enhancement with:
  • 9:16 vertical format conversion with smart cropping/blurred background
  • Audio normalization to YouTube standards (-14 LUFS)
  • Dynamic text overlays (title and captions)
  • Speed adjustment (1.35x default) with pitch preservation
  • High-quality H.264 encoding optimized for mobile viewing
• 📝 Smart Transcription - Powered by faster-whisper for accurate speech-to-text with timestamp precision
• 🚀 YouTube-Ready Metadata - Automatically generates:
  • SEO-optimized titles and descriptions (300+ characters)
  • 20-50 relevant tags for maximum discoverability
  • Upload-ready content packages with proper formatting
• 📊 Progress Tracking - Real-time progress bars for all pipeline operations
• 🎨 Multiple Video Effect Strategies - Configurable video processing approaches
• ⚙️ Highly Configurable - Comprehensive command-line arguments and environment variables

Installation

Prerequisites

• Python 3.13+
• FFmpeg installed on your system
• OpenAI API key (or compatible service like OpenRouter)

Install the Package

Clone the repository and install as a global tool:

git clone git@github.com:vitalii-honchar/youtube-shorts-creator.git
cd youtube-shorts-creator
uv tool install -e .

This creates a global shorts-creator command you can run from anywhere.

Configuration

Environment Variables

Create a .env file in your working directory with the required API key:

# Required: OpenAI API key
YOUTUBE_SHORTS_OPENAI_API_KEY=your-api-key-here

# Optional: Custom API endpoint (defaults to OpenRouter)
YOUTUBE_SHORTS_OPENAI_BASE_URL=https://openrouter.ai/api/v1

# Optional: Model selection (defaults to gpt-5-mini)
YOUTUBE_SHORTS_MODEL_NAME=openai/gpt-5-mini

# Optional: Default settings
YOUTUBE_SHORTS_SHORTS_NUMBER=3
YOUTUBE_SHORTS_SHORT_DURATION_SECONDS=60
YOUTUBE_SHORTS_SPEED_FACTOR=1.35
YOUTUBE_SHORTS_WHISPER_MODEL_SIZE=medium

Usage

Basic Usage

# Process a video with default settings
shorts-creator -v /path/to/your/video.mp4

Command Line Arguments

Argument	Short	Description	Default
--video	-v	Path to input video file	Required
--shorts	-s	Number of shorts to generate	3
--short-duration	-sd	Minimum duration of each short in seconds	60
--duration	-d	Max input video duration to process (seconds)	None (full video)
--strategy		Video effects strategy (basic)	basic
--debug		Enable debug mode with verbose logging	false
--no-refresh	-nr	Use cached files if available (opposite of refresh)	true

Note: By default, the tool refreshes/regenerates content. Use -nr to use cached files.

Examples

# Generate 3 shorts with minimum 30 seconds each
shorts-creator -v tutorial.mp4 -s 3 -sd 30

# Process only first 5 minutes of a long video
shorts-creator -v long-lecture.mp4 -d 300 -s 4

# Use cached files (faster processing)
shorts-creator -v content.mp4 -nr

# Process with custom settings and debug mode
shorts-creator -v podcast.mp4 -s 6 -sd 45 -d 1800 --debug

# Use specific video effects strategy
shorts-creator -v demo.mp4 --strategy basic --debug

Output Structure

The tool creates the following directory structure:

shorts-creator/                        # Output directory
├── extracted_audio.mp3                # Extracted audio from input video
├── speech.json                       # Transcription with precise timestamps
├── shorts.json                       # AI analysis with metadata
├── captions_0.ass                    # Generated captions for each short
├── captions_1.ass                    
├── short_0.mp4                       # Generated shorts (ready for upload)
├── short_1.mp4
└── ...

Note: The poc/ directory contains proof-of-concept experiments and research code, not production functionality.

Advanced Configuration

Custom AI Models

You can use different AI models by setting environment variables:

# Use OpenAI directly
YOUTUBE_SHORTS_OPENAI_BASE_URL=https://api.openai.com/v1
YOUTUBE_SHORTS_MODEL_NAME=gpt-4o-mini

# Use local models (if compatible with OpenAI API)
YOUTUBE_SHORTS_OPENAI_BASE_URL=http://localhost:1234/v1

Video Processing Settings

The following settings control video processing quality:

• Duration Requirements: Minimum duration enforced (can be 20-30% longer, never shorter)
• Speed Factor: 1.35 (35% faster playback while preserving pitch)
• Audio Standards: -14 LUFS loudness with -1 dBFS peak limiting
• Video Quality: H.264 High profile, optimized bitrate, 1080x1920 resolution
• Text Overlays: Auto-generated titles and subtitle captions (ASS format)
• Whisper Model: medium or large for high-accuracy transcription
• Video Effects: Configurable strategies (basic effects pipeline included)

Effects Pipeline Details

The current BASIC strategy includes:

1. Audio Normalization - YouTube-standard loudness (-14 LUFS) and peak limiting (-1 dBFS)
2. Format Conversion - Smart 9:16 aspect ratio with intelligent content-aware cropping
3. Text Overlays - AI-generated titles positioned at top of frame
4. Caption Generation - Automatic subtitle creation with precise timing (ASS format)
5. Visual Effects - Blur transitions and speed adjustments with pitch preservation
6. Professional Encoding - Optimized H.264 output for mobile consumption

Additional strategies can be implemented by extending the VideoEffectsStrategy enum.

Troubleshooting

Common Issues

• FFmpeg not found: Install FFmpeg and ensure it's in your PATH
• API key errors: Check your .env file has YOUTUBE_SHORTS_OPENAI_API_KEY
• GPU/CUDA issues: The tool uses CPU-only processing by default for compatibility
• Font rendering: Built-in fonts are included, no system fonts required

Performance Tips

• Use -d to limit processing time for very long videos
• Smaller short durations (-sd) process faster
• Consider using faster models for development/testing

License

This project is a proof-of-concept for educational and research purposes.