k4yt3x/video2x

What it does

Video2X is a video super-resolution and frame-interpolation tool. It takes a low-resolution or low-frame-rate video and produces a higher-resolution or higher-frame-rate output using machine-learning models: Anime4K v4, Real-ESRGAN, Real-CUGAN and RIFE, plus custom MPV-compatible GLSL shaders. It targets people who want to clean up or enlarge existing footage, especially animation, without manually stitching together separate model pipelines.

Version 6 is a C/C++ rewrite with a new processing pipeline, a Windows GUI and installer, and Linux packages. It has two modes: filtering (upscaling) and frame interpolation. Models run through ncnn and Vulkan, and processing claims to require no extra disk space beyond the final output README 6.0.0 notes feature details.

Quick start

For Windows, download and run the 6.4.0 Windows installer. The README also points to the latest releases and a mirror site if GitHub downloads are unavailable. After installing, open the GUI and set the input video, output path, model and mode; the desktop usage guide covers the interface.

For Linux, use an AUR package such as video2x-qt6, the archlinuxcn packages, or the Video2X-x86_64.AppImage from the releases page. A container image is also available for Linux and macOS, with instructions in the container guide.

To check that a setup works, the README provides a 240P standard test clip and matching upscaled samples for Real-CUGAN, Real-ESRGAN and waifu2x test clip section.

Requirements

Precompiled binaries require an AVX2-capable CPU (Intel Haswell or newer, AMD Excavator or newer) and a Vulkan-capable GPU. The stated minimums are NVIDIA Kepler, AMD GCN 1.0, and Intel HD Graphics 4000 hardware requirements. Upscaling is GPU-intensive, so older hardware will limit practical throughput. Output resolution and frame-rate gains also depend on the selected model and settings.

If you do not have a suitable GPU, the project offers a Google Colab notebook with a free GPU session limited to 12 hours; the README asks users not to run it continuously Colab section. Video2X is licensed under AGPL-3.0; bundled or depended-on components such as FFmpeg, ncnn, Anime4K, Real-ESRGAN, Real-CUGAN and RIFE carry their own licenses license section.

HotRepo's take

The only recorded daily snapshot is from 2026-11-05 on the C++ language list, where the project ranked 6th and gained 51 stars that day. That single observation shows continued attention around the 6.x C++ release, but it is not enough to establish a longer trend. For readers, the relevant split is between the GUI installer for Windows and the package/container routes on Linux: the project supplies real build, install and usage documentation rather than only download links, and its model choices map to common anime upscaling and frame-interpolation workflows.