Show HN: Local-First AI Video Upscaler with CPU Fallback

[pratik227]

I built a Python-based video upscaler that prioritizes privacy and local execution.

Most AI video tools today are cloud-based, which isn't ideal for private media or long-form archival projects. This tool brings a professional upscaling pipeline directly to your machine.

Why I built this: I wanted a simple "one-command" way to take old SD footage (like home movies or archival clips) and upscale them to 1080p and 4K without complex configurations or data leavings my hardware.

Key Features: - Hybrid Support: Auto-detects NVIDIA CUDA or Apple Silicon (MPS). - Smart Fallback: If no GPU is found, it automatically switches to OpenCV FSRCNN so it still works (though slower) on any CPU. - The Pipeline: It doesn't just "stretch" video. It uses Real-ESRGAN for AI enhancement up to 1080p, then applies high-quality Lanczos interpolation for 4K delivery. - Format Handling: Handles 4:3 to 16:9 aspect ratio correction with smart padding and preserves/remuxes the original audio track.

GitHub: https://github.com/pratik227/upscale_video_4k

Sample Output - https://github.com/pratik227/upscale_video_4k/blob/main/outp...