OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Infrastructure & Serving

Vs Mlrt

Vs Mlrt is an open-source collection of VapourSynth machine-learning filter runtimes for CPU and GPU inference, with plugins based on OpenVINO, ONNX Runtime, TensorRT, TensorRT-RTX, MIGraphX and NCNN/Vulkan, plus a Python wrapper for selecting bundled models and backends.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Vs Mlrt — OpenViST project profile in AI Infrastructure & Serving.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Vs Mlrt when a VapourSynth workflow needs configurable machine-learning inference across supported CPU or GPU backends, especially when hardware-specific plugins or the unified Python wrapper fit the processing setup.

Who it is for

  • VapourSynth users building AI-assisted video-processing workflows.
  • Developers and power users targeting different CPU, GPU or Apple SoC inference backends.
  • Users working with supported upscaling, restoration or frame-processing models.

Documented uses

  • Run supported machine-learning filters inside VapourSynth.
  • Upscale video with bundled models such as waifu2x, RealESRGAN or Real-CUGAN.
  • Restore frames with DPIR, SCUNet or ArtCNN.
  • Select CPU or GPU execution through the available runtimes and unified Python wrapper.
  • Use ONNX models with NCNN/Vulkan through on-the-fly conversion to NCNN format.
Before you choose

Requirements & limits

The project is publicly available as a GPL-3.0 repository. Its README describes release installation by extracting plugin files into the VapourSynth plugins directory.

  • The target hardware determines which runtime is relevant: OpenVINO targets x86 CPUs and Intel GPUs, ONNX Runtime supports CPU and CUDA GPU execution, and other plugins target NVIDIA, AMD or Vulkan-capable GPUs.
  • TensorRT and MIGraphX require building an engine on the machine where the filter will be used.
  • NCNN/Vulkan offers a device-independent Vulkan path and on-the-fly ONNX conversion, but the project identifies it as slower than the CUDA-based alternatives described there.
  • The unified vsmlrt.py wrapper can select among bundled models and backends through one interface.
  • The workflow depends on VapourSynth and on the models and usage patterns supported by the project documentation.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install a vs-mlrt VapourSynth plugin

Place a selected runtime plugin in VapourSynth’s plugins directory.

Before you start

  • A VapourSynth installation and its plugins directory.
  • A vs-mlrt runtime release matching the intended backend and hardware.

Steps

  1. Choose a runtime based on the target hardware and backend described in the project README.

  2. Download the latest release for that runtime.

  3. Extract the release files into the VapourSynth plugins directory.

  4. Consult the runtime-specific project directory for further details.

Expected outcome

The runtime’s release files are placed in VapourSynth’s plugins directory; further runtime-specific setup is described in its project directory.

Guide limitations

  • The README does not provide runtime-specific configuration or model-execution steps.
  • TensorRT requires an additional step to build an engine from an ONNX network on the machine that will use the filter.
  • NCNN/Vulkan is described as slower than other GPU runtimes.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
4 October 2026
Original website title
AmusementClub/vs-mlrt · Official homepage title · retained in its source language
How we review sourcesSuggest a correction