VTKStreaming

This module provides classes to encode and stream frames from a VTK OpenGL render window using video codecs. It supports video encoding with VP9 (through libvpx) and H.264/H.265 (through NVENC).

Installation

VTKStreaming is available on PyPi on the following platforms:

• Linux x86_64 for python 3.10 to 3.13 included.
• Windows x86_64 for python 3.10 to 3.13 included.
• MacOSX arm64 for python 3.10 to 3.13 included.

It is currently based on VTK 9.6.0.

pip install vtk-streaming

Building from source

There are two build flows:

• Development build: an editable install into a local virtual environment, built without build isolation. Rebuilds are fast and the generated compile_commands.json stays valid, so clangd/IDE tooling works.
• Release build: cibuildwheel produces the exact same distributable wheel as CI, in an isolated environment.

Requirements (both flows):

• Linux: A C++ compiler (GCC 11.4+ or any other compiler supported by VTK.)
• Windows: MSVC (Visual Studio Build Tools). (ensure visual studio environment is initialized)
• macOS: Xcode command line tools. (xcode-select --install should have completed successfully)

We recommend uv; it reads the vtk-sdk package index from pyproject.toml, so no extra index flags are needed.

Development build

The editable install runs with --no-build-isolation, so the build requirements (scikit-build-core, vtk-sdk, vtk-sdk-python-wheel-helper) must be present in the environment first, the dev dependency group installs them along with the test dependencies.

Linux/macOS:

uv venv -p 3.13
. .venv/bin/activate
uv pip install --group dev
uv pip install -e . --no-build-isolation

Windows (open a PowerShell with MSVC initialized, e.g. Visual Studio Developer PowerShell):

uv venv -p 3.13
.venv\Scripts\Activate.ps1
uv pip install --group dev
uv pip install -e . --no-build-isolation

After changing C++ sources, rebuild by re-running the editable install:

uv pip install -e . --no-build-isolation

The CMake build tree persists under build/<wheel_tag>/, which keeps generated module headers on disk and compile_commands.json valid.

Run the tests with:

pytest tests/ -v

Release build (reproduce CI artifacts)

Wheels are built with cibuildwheel in an isolated environment; all configuration lives in [tool.cibuildwheel] in pyproject.toml.

Requirements:

• Linux: Docker (the build runs in a manylinux container).
• Windows: MSVC (Visual Studio Build Tools). (ensure visual studio environment is initialized)
• macOS: Xcode command line tools. (xcode-select --install should have completed successfully)

Build a wheel for one Python/platform target:

# Linux
uvx cibuildwheel --only cp310-manylinux_x86_64

# Windows
uvx cibuildwheel --only cp310-win_amd64

# macOS
uvx cibuildwheel --only cp310-macosx_arm64

uvx comes with uv; alternatively pipx run cibuildwheel or if you use pip:

pip install cibuildwheel

# Linux
cibuildwheel --only cp310-manylinux_x86_64

# Windows
cibuildwheel --only cp310-win_amd64

# macOS
cibuildwheel --only cp310-macosx_arm64

Substitute cp310/cp311/cp312/cp313 to target other Python versions. The wheel is written to wheelhouse/ and can be installed directly:

pip install wheelhouse/vtk_streaming-*.whl

Examples

1. examples/simple_encoder_decoder.py - Live VP9 encode/decode round-trip with two render windows side by side.
2. examples/resize_encoder_decoder.py - VP9 encode/decode round-trip that survives window resizes.
3. examples/simple_nvenc_record.py - Record a render window for later playback using NVENC. This needs ffplay to playback the .h264 file.

Getting help

• For issues with the API, usage or bugs in VTKStreaming libraries, please report them on the original repository.
• For issues with wheel installation, supported Python and VTK versions or Python-side issues, please report them on the GitHub Fork.