🎥 AI Video Enhancer & Upscaler

A local-first, privacy-friendly AI video upscaling tool built in Python. Upscale videos to 1080p and 4K using Real-ESRGAN (AI) with automatic fallback to FSRCNN (CPU).

No cloud. No uploads. Fully offline.

---

✨ Features

AI Upscaling

• Real-ESRGAN for high-quality AI enhancement
• NVIDIA CUDA support
• Apple Silicon (MPS) support

Smart Fallback

• Automatically switches to OpenCV FSRCNN
• Works on CPU-only systems without configuration

1080p → 4K Pipeline

• AI upscale to 1080p
• High-quality Lanczos interpolation for 4K delivery

Aspect Ratio Handling

• Automatic 4:3 → 16:9 correction
• Smart padding (no stretching or cropping)

Audio Preservation

• Extracts original audio
• Lossless remux into upscaled video

Local & Offline

• No uploads
• No server-side processing
• No data leaves your machine

---

🖥 Platform & Hardware Support

Operating Systems

• macOS (Intel & Apple Silicon)
• Linux
• Windows

Acceleration Backends

Hardware	Backend	Status
NVIDIA GPU	CUDA	Real-ESRGAN
Apple Silicon (M1–M3)	MPS	Real-ESRGAN
CPU-only	OpenCV FSRCNN	Fallback

---

🚀 Getting Started

Prerequisites

• Python 3.8+
• FFmpeg (must be available in PATH)

---

Install FFmpeg

macOS

brew install ffmpeg

Linux

sudo apt install ffmpeg

Windows

Download from: https://ffmpeg.org/download.html Ensure ffmpeg is added to PATH.

---

📦 Installation

Clone the repository

git clone https://github.com/pratik227/video-enhance.git
cd video-enhance

Install dependencies

pip install -r requirements.txt

Download model weights

AI Upscaling (Recommended) Place one of the following files in the project root:

• RealESRGAN_x4plus.pth
• RealESRGAN_x4.pth

CPU Fallback Place this file in the models/ directory:

• FSRCNN_x3.pb

---

🛠 Usage

Rename your input video to match the default name OR update the INPUT_VIDEO variable in the script.

python upscale_video.py

Output

• AI-enhanced 1080p video
• High-quality 4K video
• Original audio preserved

---

📚 Python Dependencies

• torch
• opencv-python
• ffmpeg-python
• numpy
• tqdm
• realesrgan
• ai-forever Real-ESRGAN

---

⚙️ Processing Notes

• GPU acceleration is auto-detected at runtime
• FSRCNN fallback is used automatically if GPU is unavailable
• Optimized for long-form and archival video processing

Suitable for

• Old SD footage
• Personal media restoration
• Content remastering
• Offline processing pipelines

---

💖 Support & Sponsorship

If this project helps you, consider supporting its development.

💝 GitHub Sponsors https://github.com/sponsors/pratik227

☕ Buy Me a Coffee https://buymeacoffee.com/pratik227

⭐ Free Support

• Star the repository
• Share on X / Reddit / Hacker News
• Recommend it to others

---

🧑‍💻 Maintainer

Pratik Patel Independent builder focused on local-first, privacy-respecting tools

GitHub: https://github.com/pratik227

---

📄 License

MIT License See the LICENSE file for details.