README.md

---
title: Zero2x Video Upscaler
emoji: 🎞️
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: mit
hardware: t4-small
---

# 🎞️ Zero2x AI Video Upscaler

Professional AI-powered video upscaling using state-of-the-art deep learning models with advanced region detection and motion analysis.

## ✨ Features

- 🚀 **Multiple AI Models**: RealESRGAN, SwinIR, CUGAN for different video types
- ⚡ **GPU Acceleration**: Optimized for Hugging Face Spaces GPU
- 🎬 **Advanced Processing**: Smart region detection with SAD algorithm
- 🔍 **Motion Detection**: ORB-based motion analysis for optimal frame processing
- 🧠 **Intelligent Upscaling**: Hybrid full-frame and region-based processing
- 🎯 **Memory Efficient**: Optimized CUDA memory management

## 🧑‍🔬 AI Models Supported

- **RealESRGAN x4+**: Best for realistic videos and photos (4x upscaling)
- **RealESRGAN x2+**: Faster processing for moderate enhancement (2x upscaling)
- **RealESRGAN Anime**: Specialized for anime and cartoon content
- **SwinIR 4x**: Advanced transformer-based upscaling for maximum detail
- **CUGAN**: Optimized for animated content upscaling

## 🚀 How to Use

1. **Upload Video**: Drag and drop or click to upload any video file
2. **Choose Model**: Select the best AI model for your content type
3. **Configure Settings**: Adjust advanced parameters for optimal results
4. **Process**: Click "Upscale Video" and wait for the AI enhancement
5. **Download**: Save your enhanced high-resolution video

## ⚙️ Advanced Features

- **Region Detection**: Only upscale changed areas for faster processing
- **Motion Analysis**: Automatic detection of camera movement and scene changes
- **Similarity Tracking**: Skip identical frames to save processing time
- **Segmented Processing**: Break videos into manageable segments
- **Full-Frame Fallback**: Automatic switching between region and full-frame modes

## 💡 Tips for Best Results

- **Anime/Cartoons**: Use RealESRGAN Anime models
- **Real Videos**: Use RealESRGAN x4+ for maximum quality
- **Speed vs Quality**: Use x2+ models for faster processing
- **Enable Regions**: Turn on region detection for significantly faster processing
- **Large Videos**: The system handles high-resolution inputs efficiently

## 🛠️ Technical Details

- **Framework**: Gradio + PyTorch + OpenCV
- **Algorithms**: Custom SAD (Segmented Absolute Differences) + ORB motion detection
- **Models**: Spandrel model loading system for universal compatibility
- **GPU**: CUDA-accelerated processing with automatic memory management
- **Processing**: Hybrid approach combining region-based and full-frame upscaling

## 📊 Performance

- **Memory Optimized**: Efficient GPU memory usage with automatic cleanup
- **Smart Processing**: Skips duplicate frames and processes only changed regions
- **Progress Tracking**: Real-time progress updates with time estimation
- **Batch Optimized**: Handles multiple frames efficiently

## 🔗 Links

- **Original Project**: Zero2x Video Upscaler
- **Models**: Compatible with standard upscaling model formats (.pth)
- **Documentation**: Comprehensive parameter explanations in the interface

---

*🎞️ Professional AI video upscaling with intelligent region detection and motion analysis*