Skip to content

Repository files navigation

mpv-h

Platform mpv Super Resolution VapourSynth TensorRT RIFE License Release

English | Simplified Chinese

mpv-h is a personal Windows portable mpv setup. It combines mpv configuration, uosc, thumbfast, shader presets, and a VapourSynth/vsmlrt/TensorRT/RIFE interpolation profile.

This is not an official mpv build.

mpv-h quick demo

Anime4K shader profile RIFE frame interpolation
Anime4K shader profile RIFE frame interpolation

Watch the full demo video

Features

  • Portable Windows mpv configuration.
  • uosc interface and right-click menu.
  • thumbfast thumbnail previews.
  • Anime4K and FSRCNNX shader profile toggles.
  • RIFE TensorRT interpolation toggle through VapourSynth.
  • mpv-h.exe launcher that injects the bundled VapourSynth/vsmlrt runtime into mpv's process environment.
  • Optional Windows file association install/uninstall scripts.

Requirements

  • Windows 10/11 x64.
  • NVIDIA GPU for the bundled RIFE TensorRT profile.
  • Recent NVIDIA driver. The first Release was tested with driver 596.36; NVIDIA-SMI reported CUDA compatibility 13.2.
  • 7-Zip for extracting split Release archives.

The full Release package includes the required portable runtime: mpv, FFmpeg, yt-dlp, VapourSynth, vsmlrt, TensorRT runtime DLLs, CUDA runtime DLLs required by the bundled TensorRT stack, and ONNX models. It does not redistribute NVIDIA trtexec.exe. Users do not need to install the CUDA Toolkit, VapourSynth, Python, or the .NET SDK separately for normal playback.

Download

Use the latest GitHub Release if you want a ready-to-run package. Download all archive parts into the same folder, then extract the .7z.001 file with 7-Zip. Do not rename the .001 / .002 files.

The source repository is for configuration, scripts, launcher source, and documentation. Large runtime files such as mpv.exe, FFmpeg, yt-dlp, VapourSynth, TensorRT runtime files, CUDA runtime DLLs, and ONNX models are distributed through Releases instead of Git.

RIFE TensorRT Setup

RIFE interpolation needs NVIDIA trtexec.exe to build a local TensorRT engine on first use. NVIDIA documents trtexec as a TensorRT command-line tool, but mpv-h does not redistribute it.

To enable first-time RIFE engine generation:

  1. Download the matching TensorRT Windows zip from NVIDIA: https://developer.nvidia.com/tensorrt.
  2. For this Release, the recommended NVIDIA direct link is https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.14.1/zip/TensorRT-10.14.1.48.Windows.win10.cuda-13.0.zip.
  3. Extract the NVIDIA zip and copy bin\trtexec.exe to:
mpv-h-2026.05.23-full\VapourSynth\Lib\site-packages\vapoursynth\plugins\vsmlrt-cuda\trtexec.exe

After the engine has been generated, later RIFE launches can reuse the local engine cache unless the GPU, driver, TensorRT runtime, model, precision, or input shape changes.

Usage

For the full Release package:

  1. Extract the archive.
  2. Run mpv-h.exe.
  3. Optionally run installer/mpv-h-install.bat as administrator to register mpv-h in Windows Default Apps.

To remove the Windows app registration, run:

.\installer\mpv-h-uninstall.bat

Shortcuts

Shortcut Action
Alt+1 Toggle Anime4K
Alt+2 Toggle FSRCNNX
Alt+0 Disable all GLSL shaders
Alt+R Toggle RIFE TensorRT interpolation
Alt+9 Disable video filters and restore hardware decoding
Right click Open uosc menu

Anime4K/FSRCNNX shader upscaling and RIFE TensorRT interpolation are intentionally mutually exclusive. Running both at the same time is usually too heavy for smooth video playback. The profile also switches hardware decoding by workload: shader upscaling restores hwdec=auto-safe, while RIFE uses hwdec=no because this tested machine drops frames with hardware decoding enabled during VapourSynth interpolation.

Repository Layout

portable_config/        mpv configuration, scripts, uosc resources, fonts, shaders
tools/mpv-h-launcher/   launcher source code
installer/              Windows registration scripts and icon
scripts/                maintainer build and release packaging scripts
docs/                   configuration and release maintenance notes
LICENSES/               third-party license texts

Known Assumptions

  • The RIFE profile defaults to NVIDIA device_id=0.
  • TensorRT engine files are generated locally on first use and should not be copied between machines.
  • First RIFE startup can be slow while TensorRT builds an engine for the current GPU, driver, model, precision, and input shape.
  • The default config is intentionally neutral. High-refresh-display tuning such as display-fps-override=240 is documented but not enabled by default.

See docs/configuration.md for tuning notes.

Maintainer Notes

The launcher can be rebuilt with the Windows .NET Framework C# compiler:

powershell -ExecutionPolicy Bypass -File .\scripts\Build-Launcher.ps1

The repository root is also the local runnable mpv-h directory. Runtime binaries such as mpv.exe, ffmpeg.exe, yt-dlp.exe, and VapourSynth/ live next to the tracked portable_config/ but are ignored by Git.

The full Release package can be built directly from the repository root:

powershell -ExecutionPolicy Bypass -File .\scripts\New-ReleasePackage.ps1 -Version 2026.05.23

Keep these root-level runtime files separate from any personal mpv installation. portable_config/ in this repository remains the source of mpv-h configuration.

More details are in docs/RELEASE.md.

Licensing

The original mpv-h code and configuration are licensed under GPL-3.0-or-later. Bundled third-party components keep their original licenses.

See THIRD-PARTY-NOTICES.md and LICENSES/ for component-specific licensing.

About

Personal Windows mpv portable setup with uosc, thumbfast, shader presets and a VapourSynth/vsmlrt/TensorRT/RIFE interpolation.

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages