Installation

Installation is simple. Download the latest release, stick it in a (ideally empty) folder, and it will fetch all these things:

• ChaosKing's VapourSynth Fatpack, which contains a portable version of Python, VSEdit for editing, and a ton of VapourSynth plugins and scripts, among other things.
• 7zip, for extracting archives.
• A copy of this repository.
• Some minor helper PyPI packages, like pySmartDl and OpenCV.
• Kice's MXnet VapourSynth plugin.
• Portable Subversion.
• The appropriate Python version of mxnet for your GPU/CUDA version.
• PyTorch
• The VSGAN VapourSynth script for ESRGAN PyTorch models.
• FFMPEG's latest releast
• And finally, optionally, and if you have an Nvidia GPU, the script will attempt to audodetect the version of CUDA/cuDNN installed and offer to silently install them for you. This is one of the trickiest requirements for super resolution setups. Note that the CUDA script is a bit buggy, and needs testing!

With the exception of CUDA and cuDNN, the install is totally portable. If something goes wrong, you can delete the folder the installer created, and run the installer again. Downloads are cached by pip, as well as pySmartDL in the Windows %TEMP%/PySmartDL folder, so running the installer a 2nd time shouldn't use much bandwidth.