🎥 MitoClipper Evolution

Professional AI-Powered Video Clipping & Auto-Reframing Suite

MitoClipper Evolution is a sophisticated tool designed to transform long-form content (podcasts, interviews, streams) into viral short-form clips. Using a modern stack and high-performance ML models, it automates the entire pipeline from detection to rendering and uploading.

• 🚀 Dual-Engine LLM: Blazing fast metadata generation using Groq (Llama-3) with local fallback for offline privacy.
• 🎯 Precision Face Tracking: High-accuracy auto-framing for vertical (9:16) crops powered by MediaPipe.
• ⚡ Async Pipeline: Completely rewritten in FastAPI with background task management for zero-lag UI performance.
• ✨ Professional Subtitles: Dynamic, animated ASS subtitles (Hormozi/MrBeast style) powered by the pysubs2 engine.
• 📟 Real-time Control Center: A sleek Glassmorphism UI with WebSocket log streaming and live progress tracking.
• 🎬 Parallel Rendering: Multi-threaded FFmpeg rendering for maximum performance.
• ☁️ Auto-Upload: Integrated YouTube API support for automated publishing.

• Backend: FastAPI (Python 3.10+)
• Frontend: Tailwind CSS + Vanilla JS + Lucide Icons
• ML/Computer Vision: MediaPipe, OpenAI Whisper (Faster-Whisper), PyTorch, Librosa
• Video Processing: FFmpeg, pysubs2
• Acceleration: CUDA (Nvidia), MPS (Apple Silicon), or CPU fallback

• Python 3.10+
• FFmpeg installed and in your PATH
• Groq API Key (Optional, for 10x faster metadata)

# Clone the repository
git clone https://github.com/yourusername/mitoclipper.git
cd mitoclipper

# Activate your venv
source clipenv/bin/activate

# Install dependencies
pip install -r requirements.txt

Create a .env file in the root directory:

MITOCLIPPER_GROQ_API_KEY=your_groq_key_here
MITOCLIPPER_WHISPER_MODEL_ID=base
MITOCLIPPER_LLM_DEVICE=cpu # or cuda

# Start the FastAPI server
python app/main.py

Visit http://localhost:5000 in your browser.

MitoClipper is built to be resilient.

• If MITOCLIPPER_GROQ_API_KEY is found, it will use Groq for instantaneous clip titles and descriptions.
• If no key is found, it automatically falls back to a local Qwen2.5-1.5B model.

MitoClipper features a dynamic auto-framing engine for vertical (9:16) crop generation:

• Precision Sampling: Instead of static crops, it tracks face locations across time using MediaPipe Face Detection.
• EMA Smoothing: A low-pass Exponential Moving Average (EMA) filter is applied to the raw coordinates to smooth transitions and eliminate camera jitter.
• FFmpeg Expression Compilation: The smoothed motion path is compiled into a single, time-based piecewise-linear interpolation expression passed directly to FFmpeg's crop filter, performing smooth panning between speakers at native rendering speeds.

• FFmpeg
• Python 3.10+
• MediaPipe (for Face Tracking)
• Pysubs2 (for Subtitles)
• FastAPI / Uvicorn

MIT License - Copyright (c) 2026 MitoClipper Team