How Vapourkit Works

The mental model behind Vapourkit, end to end. Read this after Process Your First Video if you want to understand what happens under the hood.

VapourSynth

VapourSynth is a Python-based frame server and video processing framework. Every operation Vapourkit performs - decoding, filtering, AI upscaling, format conversion - is expressed in a VapourSynth graph and driven by a Python script (.vpy).

Vapourkit composes that script for you. When you press Start processing, the app creates a temporary .vpy file from its processing template and hands it to the installed VapourSynth runtime. The UI is a frontend over that script.

The processing pipeline

Each run is a single ordered chain:

input video -> colorimetry setup -> filter 1 -> filter 2 -> ... -> encode -> output

• Source decoding uses BestSource on the input file.
• Colorimetry is applied before the filter chain according to the output settings.
• The AI model is a filter. It can run before or after other filters, in whatever order you set in the filter panel. You can also disable the model and run a filter-only pipeline.
• Encoding happens after the chain completes; the final clip is converted to the selected format and written to the chosen container. See File Formats.

Backends by operating system

Vapourkit supports three inference backends. The backend picker only shows options supported by the current platform.

Platform	Available backends	Default
Windows	TensorRT, DirectML, NCNN Vulkan	TensorRT
Linux	NCNN Vulkan; TensorRT on NVIDIA with a compatible CUDA/TensorRT stack	NCNN Vulkan

DirectML is Windows-only.

• TensorRT (TRT) is NVIDIA’s optimizing inference runtime. Models are compiled into .engine files targeted to the GPU. The first build can take several minutes, but inference is usually fastest afterward. TensorRT is available on Windows and Linux, but Linux requires a compatible NVIDIA CUDA/TensorRT installation.
• DirectML (DML) is a Windows-native ONNX Runtime backend. It runs ONNX models directly on AMD, Intel, and NVIDIA GPUs with DirectX 12 support, without an engine build step.
• NCNN Vulkan runs ONNX models directly through Vulkan. It is available on Windows and Linux and requires a working Vulkan loader and GPU driver. It is the Linux default and provides the cross-vendor Linux path.

Models: VSR vs image-based

Vapourkit ships two kinds of AI upscaling model.

• VSR (Video Super-Resolution) models are temporally aware: each output frame uses several neighbouring source frames, which improves stability over time.
• Image-based models process each frame independently. They still work on video, but can show frame-to-frame shimmer or flicker on fine detail.

For the bundled model list, see Choose a Model. For importing your own, see Use Your Own Model.

Filters

A filter in Vapourkit is a small piece of VapourSynth Python with a simple contract: it receives a variable named clip and reassigns clip to its output.

# CAS Sharpen.vkfilter
clip = core.cas.CAS(clip, sharpness=0.5, planes=0)

Filters are persisted as .vkfilter files - TOML wrappers with a name, category, description, and code block. You can mix filters freely, reorder them, or write your own.

Templates vs workflows

Two save formats serve different scopes:

• Filter templates (.vkfilter) - one filter saved for reuse.
• Workflows (.vkworkflow) - the complete pipeline, including enabled filters, the selected model, output settings, and segment selection.

See Save & Reuse Workflows for the current Open, Import, and Export flow.

Single video vs queue

Both flows build and run the same kind of pipeline. The difference is when the configuration is captured:

• Single video uses the filters, model, and settings currently selected in the UI.
• Queue stores a workflow snapshot for each item when it is added. Editing the UI afterwards does not change that queued item. See Process Multiple Videos.

Next

• Add Filters - change the processing result
• Save & Reuse Workflows - repeat a setup
• Process Multiple Videos - use the queue
• Filter Reference - every bundled filter