CorridorKey: Native AI Keying Runtime for DaVinci Resolve & OFX Pipelines

Dual-Mode Deployment

The runtime operates in two distinct modes, both leveraging the same quantized ONNX inference engine:

Hardware Abstraction Layer

Unlike Python-based alternatives, CorridorKey compiles to native binaries with direct GPU compute:

• Apple Silicon: Metal Performance Shaders (MPS) backend, optimized for unified memory architecture
• NVIDIA RTX: CUDA 12 + TensorRT 8.6 for FP16 inference
• Memory Footprint: ~3.8GB VRAM for 4K UHD processing via aggressive model quantization

Configuration & Extensibility

Keying parameters (spill suppression strength, edge detail preservation) are exposed via JSON schemas in CLI mode, while the OFX plugin provides standard color page controls. The runtime supports FFmpeg piping for custom input pipelines.

Solving the "Cloud Round-Trip" Tax

Traditional AI keying requires uploading footage to Runway or similar services, waiting for processing, then downloading—breaking frame-accurate editorial workflows. CorridorKey performs inference locally at 30-60fps, allowing colorists to adjust keys while scrubbing through timeline playback.

"The spill suppression on motion-blurred green screens handles edges that make standard Delta Keyers choke—this is the first AI keyer that feels like a native Resolve tool."

Edge Case Handling

Where traditional chroma keyers fail on hair detail, transparent fabrics, or uneven lighting, the underlying segmentation model (likely a modified Deep Image Matting or custom U-Net variant) provides alpha channel granularity. The adaptive spill suppression algorithm runs as a post-processing pass on the GPU, avoiding the color casts common in simple despill operations.

Pipeline Integration

The CLI mode exposes a --json-metadata flag for render managers like Deadline or Tractor, returning per-frame confidence scores and processing timestamps. This enables automated quality control—flagging frames with low confidence for manual review rather than failing entire renders.

Throughput Benchmarks

Based on reported specs and typical ONNX Runtime performance on similar segmentation models:

Comparative Analysis

The Trade-off: CorridorKey sacrifices the absolute pixel-perfect accuracy of research-grade matting models for interactive performance. For broadcast delivery and YouTube content, the quality delta is negligible; for feature film VFX, it serves as a rapid blocking tool before handing to roto artists.

OFX Host Compatibility

Built against the OFX 1.4 specification, the plugin targets:

• DaVinci Resolve Studio 18.6+ (primary target, tested on macOS and Windows)
• Nuke 14+ (via OFX, limited testing)
• Baselight (theoretical support, unverified)

Note: The free version of Resolve does not support OFX plugins—this requires the Studio license ($295), positioning CorridorKey firmly in the professional tier.

Automation & CLI Ecosystem

The CLI binary supports stdin/stdout piping with FFmpeg, enabling workflows like:

Python bindings are marked as "planned" in the roadmap, which would unlock integration with OpenTimelineIO and Ftrack/ShotGrid automation scripts.

Adoption Signals

Backed by Corridor Digital (10M+ YouTube subscribers, professional VFX production house), the tool carries immediate credibility with content creators. Early GitHub activity shows forks from users at major streaming platforms and boutique post-houses. The 8 forks (high ratio to stars) suggest developers are actively studying the C++ implementation for custom VFX tool development rather than just casual use.