k4yt3x/video2x

Video2X is a tool that uses machine learning to improve the quality of videos in two ways: upscaling (increasing resolution) and frame interpolation (adding extra frames to make motion smoother). Upscaling takes a low-resolution video, say, 360p or 480p, and uses trained neural networks to intelligently fill in detail, producing a sharper 4K output that looks significantly better than simple image enlargement. Frame interpolation takes a 30 frames-per-second video and generates in-between frames to produce 60 or 120 FPS output with noticeably smoother motion. The underlying AI models it uses include Real-ESRGAN and Real-CUGAN for upscaling (which produce particularly good results on anime content), Anime4K for GPU-accelerated real-time upscaling via GLSL shaders, and RIFE for frame interpolation. All GPU processing is done through Vulkan, which means it works on NVIDIA, AMD, and Intel GPUs, any GPU that supports Vulkan. The core application is written in C++ for performance and is available for Windows and Linux. Windows users get an installer with a graphical interface in multiple languages, Linux users can install through AUR packages or use a universal AppImage. A Docker container image is also available. For users without a powerful GPU, Video2X can be run on Google Colab for free using Google's cloud GPUs. Demo videos in the README show examples of anime films upscaled from 360p to 4K and from 240p 30fps to 1080p 60fps. You would use this tool to restore and enhance older videos, improve streaming quality of compressed content, or prepare content for display on high-resolution screens.