Native AI keying runtime and OFX plugin for DaVinci Resolve, built in collaboration with Corridor Digital.

CorridorKey-Runtime provides two supported surfaces:

• DaVinci Resolve OFX plugin for interactive keying inside Resolve
• CLI runtime for local processing, diagnostics, and automation

Current public builds support:

• Windows (NVIDIA RTX)
• macOS (Apple Silicon)

• Download the latest release: Releases
• Check supported platforms: Support Matrix
• Troubleshooting and setup help: User Help
• Watch a community tutorial: Tutorial by Heiko (@hethfilms)

• Native local execution
• Practical deployment in real editing workflows
• Support for NVIDIA RTX and Apple Silicon
• Stable surfaces for both DaVinci Resolve users and CLI/integration workflows

Use CorridorKey directly inside DaVinci Resolve through the OFX plugin.

Use the CLI for:

• environment diagnostics
• scripted processing
• machine-readable JSON output
• automation and integration workflows

• OFX Panel Guide
• DaVinci Resolve Tutorials
• Support Matrix
• Troubleshooting

• Architecture
• Frontend
• Guidelines
• Release Guidelines
• Spec
• Contributing

Pre-packaged releases are available on the Releases page. Download the package that matches your platform and product track.

For supported hardware configurations and explicit support status per platform, see Support Matrix.

1. Download the .pkg Apple Silicon installer.
2. Run the installer with DaVinci Resolve closed.
3. Open DaVinci Resolve 20, go to the Color or Fusion page, and search for "CorridorKey" in the OpenFX Library.
4. Drag the node onto your clip. The plugin uses the MLX-accelerated path automatically on M-series chips.

1. Download the .exe installer for your hardware path:
   • Windows RTX - official Windows RTX installer for NVIDIA RTX 30 series and newer. This track ships the public FP16 quality ladder: Draft (512), High (1024), Ultra (1536), and Maximum (2048). Auto respects the safe VRAM ceiling for the active GPU tier, while manual fixed quality may attempt a higher packaged rung with explicit runtime fallback if it fails.
   • DirectML package - experimental Windows track for DirectX 12 GPUs outside the official RTX path. This track is not broadly validated across AMD, Intel, or RTX 20 series hardware and should only appear in releases when it is published intentionally.
2. Run the installer as Administrator with DaVinci Resolve closed.
3. Open DaVinci Resolve 20, go to the Color or Fusion page, and search for "CorridorKey" in the OpenFX Library.
4. Drag the node onto your clip. TensorRT RTX compilation on the first frame may take 10-30 seconds.

For plugin discovery issues, version mismatches, or unsupported hardware behavior, see Troubleshooting.

Download the portable runtime release for your platform.

• On macOS bundles and source builds, use corridorkey
• In the Windows portable runtime bundle, use ck-engine.exe

The examples below use the macOS and source-build command name corridorkey. In the Windows portable runtime bundle, replace it with ck-engine.exe.

Check hardware capability:

corridorkey doctor

Process a video with hardware-aware defaults:

corridorkey process input.mp4 output.mp4

Process with a specific preset:

corridorkey process input.mp4 output.mp4 --preset max

Process with an external Alpha Hint:

corridorkey process input.mp4 output.mp4 --alpha-hint hint.mp4

Append --json to any command to receive NDJSON event streams for pipeline integration.

• Tutorial by Heiko (@hethfilms): practical walkthrough of the CorridorKey Resolve OFX plugin, including installation and real professional workflow usage. Watch here
• Tutorial by Heiko (@hethfilms): CORRIDORKEY - How to key difficult shots. Watch here

• C++20 compiler: Visual Studio 2022 (v17.4+), Apple Clang 15+, or GCC 12+
• CMake 3.28+
• Ninja
• vcpkg with VCPKG_ROOT set

git clone https://github.com/alexandremendoncaalvaro/CorridorKey-Runtime.git
cd CorridorKey-Runtime
export VCPKG_ROOT="$HOME/vcpkg"
cmake --preset release
cmake --build --preset release

On Windows, use .\scripts\windows.ps1 -Task build -Preset release for local builds and .\scripts\windows.ps1 -Task release -Version X.Y.Z for official Windows release packaging. That canonical release command emits the official Windows RTX installer by default. Publish the experimental DirectML track only when you request it explicitly with -Track dml or -Track all.

Lower-level Windows scripts exist only as internal delegates for debugging the wrapper itself. If you invoke CMake directly, activate the MSVC developer environment first. Windows distribution artifacts include model_inventory.json and bundle_validation.json when packaging succeeds with a partial model set, so missing packaged models are explicit and do not silently change runtime behavior.

For local Windows workflow, the canonical wrapper exposes three different levels of operation:

• .\scripts\windows.ps1 -Task build -Preset release

  • build only

• .\scripts\windows.ps1 -Task package-ofx -Version X.Y.Z -Track rtx

  • package the Windows RTX installer from an already certified Windows RTX model set

• .\scripts\windows.ps1 -Task certify-rtx-artifacts -Version X.Y.Z

  • certify an already existing Windows RTX model set and write the artifact manifest without regenerating the .onnx files from the checkpoint

• .\scripts\windows.ps1 -Task regen-rtx-release -Version X.Y.Z

  • regenerate ONNX artifacts from the checkpoint, certify the RTX ladder, write the artifact manifest, and then package the Windows RTX installer

package-ofx for Windows RTX is intentionally strict. It no longer accepts a raw models\ folder by itself. The command requires a certified artifact_manifest.json that matches the packaged RTX model and *_ctx.onnx files exactly. If you only have stale or manually copied models, use certify-rtx-artifacts or regen-rtx-release first.

See CONTRIBUTING.md for development setup and PR process.

CC BY-NC-SA 4.0

You may use this software to process commercial video. You may not repackage or sell the software itself or offer it as a paid service.

• Original CorridorKey: github.com/nikopueringer/CorridorKey
• EZ-CorridorKey: github.com/edenaion/EZ-CorridorKey
• CorridorKey-Engine: github.com/99oblivius/CorridorKey-Engine
• ONNX Runtime by Microsoft
• OpenEXR by Academy Software Foundation