video2x README

The project's README from github.com/k4yt3x/video2x. As one markdown file: README.md.

Video2X: A machine learning-based video super resolution and frame interpolation framework.

🌟 Version 6.0.0

Video2X 6.0.0 highlights:

Click to see more details

Version 6.0.0 is a complete rewrite of this project in C/C++. It:

  • genuinely works this time, with much less hassle compared to the 5.0.0 beta;
  • is blazing fast, thanks to the new optimized pipeline and the efficiency of C/C++;
  • is cross-platform, available now for both Windows and Linux;
  • offers significantly better output quality with Anime4K v4, Real-ESRGAN, Real-CUGAN, and RIFE;
  • supports two modes: filtering (upscaling) and frame interpolation;
  • supports Anime4K v4 and all custom MPV-compatible GLSL shaders;
  • supports Real-ESRGAN, Real-CUGAN, and RIFE (all models) via ncnn and Vulkan;
  • requires zero additional disk space during processing, just space for the final output.

6.4.0-screenshot

🖥️ Hardware Requirements

Your system must meet the minimum hardware requirements below to run Video2X.

🪟 Install on Windows

Download the Latest Windows Installer Executable (6.4.0)

You can download the latest Windows release on the releases page. For basic GUI usage, refer to the documentation. If you're unable to download directly from GitHub, try the mirror site. The GUI currently supports the following languages:

🐧 Install on Linux

Video2X packages are available for the Linux distros listed below. A universal AppImage is also available for other distros. If you'd like to build it from source code, refer to the PKGBUILD file for a general overview of the required dependencies and commands.

📦 Container Image

Video2X container images are available on the GitHub Container Registry for easy deployment on Linux and macOS. If you already have Docker/Podman installed, only one command is needed to start upscaling a video. For more information on how to use Video2X's Docker image, please refer to the documentation.

📔 Google Colab

You can use Video2X on Google Colab for free if you don't have a powerful GPU of your own. You can borrow a powerful GPU (NVIDIA T4, L4, or A100) on Google's server for free for a maximum of 12 hours per session. Please use the free resource fairly and do not create sessions back-to-back and run upscaling 24/7. This might result in you getting banned. You can get Colab Pro/Pro+ if you'd like to use better GPUs and get longer runtimes. Usage instructions are embedded in the Colab Notebook.

💬 Telegram Discussion Group

Join our Telegram discussion group to ask any questions you have about Video2X, chat directly with the developers, or discuss super resolution, frame interpolation technologies, or the future of Video2X in general.

📖 Documentation

Comprehensive documentation for Video2X is available at https://docs.video2x.org/. It offers detailed instructions on how to build, install, use, and develop with this program.

📽️ Video Demos (Outdated)

Spirited Away Demo
Upscale demo: Spirited Away's movie trailer

Standard Test Clip

The following clip can be used to test if your setup works properly. This is also the standard clip used for running performance benchmarks.

The original clip came from the anime "さくら荘のペットな彼女."
Copyright of this clip belongs to 株式会社アニプレックス.

⚖️ License

This project is licensed under GNU AGPL version 3.
Copyright (C) 2018-2025 K4YT3X and contributors.

AGPLv3

This project includes or depends on these following projects:

Project License
FFmpeg/FFmpeg LGPLv2.1, GPLv2
Tencent/ncnn BSD 3-Clause
bloc97/Anime4K MIT License
nihui/realcugan-ncnn-vulkan MIT License
nihui/rife-ncnn-vulkan MIT License
xinntao/Real-ESRGAN-ncnn-vulkan MIT License

More licensing information can be found in the NOTICE file.

🌺 Special Thanks

Special thanks to the following individuals for their significant contributions to the project, listed in alphabetical order.

video2x (k4yt3x/video2x)

anime4k
Searching Database…

22.3K ·GitHub
What is this?Video2X is an open-source video super resolution and frame interpolation tool: it uses machine learning to upscale low-resolution videos to higher quality and to add interpolated frames for smoother motion.
Is this a desktop app, or do I run it from the command line?Both sides are covered: on Windows it ships as a GUI app with a standard installer, on Linux you get packages or an AppImage, and a Docker image covers headless runs on Linux and macOS.
Will my computer be able to run it?You need a CPU with AVX2 support (Intel Haswell 2013 or newer, AMD Excavator 2015 or newer) and a Vulkan-capable GPU — anything from NVIDIA GTX 600 series, AMD HD 7000 series or Intel HD 4000 onwards.
Does it work on macOS?There is no native macOS build; macOS is supported through the Video2X container image on the GitHub Container Registry, alongside Linux.
Is it free, and what does the license allow?It is free and open source under GNU AGPL v3 — fine for personal use, but modifications and derived projects must be released under the same license.
What if I don't have a powerful GPU?You can run Video2X on Google Colab for free: borrow an NVIDIA T4, L4 or A100 for up to 12 hours per session, but don't chain sessions back-to-back to upscale 24/7 or Google may ban your account.
Which AI models does it support for upscaling?For upscaling it ships Anime4K v4, Real-ESRGAN and Real-CUGAN, and for frame interpolation RIFE — all running through ncnn on Vulkan. It also accepts any custom MPV-compatible GLSL shaders.
Can it fix choppy low-framerate video, not just upscale it?Yes — alongside filtering (upscaling) there is a dedicated frame interpolation mode powered by RIFE, so you can raise a clip's frame rate as well as its resolution.
Does it need a lot of free disk space while processing?No. The 6.x pipeline requires zero additional disk space during processing — you only need room for the final output file.
Why was the whole project rewritten for version 6?The author rewrote Video2X from scratch in C/C++ so it genuinely works with much less hassle than the 5.0.0 beta, runs blazingly fast thanks to an optimized pipeline, and produces significantly better output quality.
Creator
k4yt3x
Size
560.8 MB
Stars
22.3K
Age
8y
Commits · 90 Days