What it does CorridorKey-Runtime is a C++ AI keying engine that ships as an OFX plugin for DaVinci Resolve and Foundry Nuke, plus a CLI and a Tauri-based GUI. It performs chroma-keying with machine learning models running natively on your hardware—no cloud round-trips, no subscription tiers. Install the OFX bundle and the corridorkey CLI lands on your PATH automatically, ready for scripting or diagnostics.

The interesting bit The project treats the OFX plugin as the primary surface but quietly makes the CLI a first-class citizen: same runtime, JSON-emitting NDJSON streams, automation-friendly. The Windows RTX track even bakes in a “quality ladder” (Draft 512 → Maximum 2048) with VRAM-aware auto-selection, so the plugin doesn’t blindly OOM your GPU. On Apple Silicon it routes through MLX automatically.

Key highlights

• OFX 1.4-compliant plugin for Resolve 20 and Nuke; host-agnostic bundle placement
• CLI with doctor hardware checks, preset quality ladders, and --json for pipeline integration
• Tauri GUI distributed separately, self-contained, no OFX install required
• Windows RTX path uses TensorRT with explicit FP16 rungs; DirectML track exists but is experimental and not broadly validated
• macOS Apple Silicon path uses MLX acceleration on M-series chips
• CC BY-NC-SA 4.0: you can use it commercially on your own footage, you cannot resell the software or offer it as a paid service

Caveats

• Windows RTX 30-series and newer only for the official RTX track; RTX 20 series and non-NVIDIA GPUs fall back to an experimental DirectML path with limited validation
• First-frame TensorRT compilation can stall 10–30 seconds in Resolve
• Source builds require C++20, CMake 3.28+, Ninja, and vcpkg; Windows release packaging is gated behind a strict artifact certification step that rejects raw model folders

Verdict Resolve and Nuke compositors who want local AI keying without leaving their timeline should grab the release installer. Pipeline engineers and automation-minded users get unexpected value from the CLI. Skip it if you’re on AMD/Intel GPUs hoping for a polished experience, or if you need to redistribute keying as a service.