Machine Learning🔨 Personal Project
AI-Powered YouTube Shorts Generator
Automated system transforming long-form YouTube videos into viral Shorts using AI transcription (Whisper), GPT-4o-mini content analysis, MediaPipe face detection, and FFmpeg processing with 98% cost reduction vs cloud alternatives.
Project Overview
🎯
Objective
Automate creation of YouTube Shorts from long videos using AI for content analysis, smart cropping, and multi-speaker detection to maximize viral potential
💼
My Role
ML Engineer & Developer - implemented end-to-end pipeline with AI integration, video processing, and cost optimization
⏱️
Timeline
1 month
🛠️
Tech Stack
Python, OpenAI Whisper, GPT-4o-mini
📈
Key Results
- ✓Complete automation: Download → Transcribe (Whisper) → Analyze (GPT-4o-mini) → Smart crop (MediaPipe) → Export (9:16 1080x1920)
- ✓Advanced features: Auto camera switching, active speaker detection, auto subtitles, watermark overlay, background music mixing
- ✓Cost optimization: $0.002/video (local Whisper medium + GPT-4o-mini) vs $0.15-0.30 with cloud services
- ✓Performance: 10min video → 2-3min processing → 3-5 clips output, 30min video → 5-10 clips
Impact
$0.002-0.005 per video
Value
98% cost reduction
📸
Visual Documentation
Screenshot 1
Screenshot 2
Screenshot 3
🔗Project Links
GitHub RepositoryPrivate
Tools & Technologies
PythonOpenAI WhisperGPT-4o-miniMediaPipeFFmpegLight-ASD