corridorkey-green-screen

What This Skill Does

CorridorKey is a sophisticated AI-powered green screen keying utility designed to address the complex problem of color unmixing in visual effects. Traditional green screen keying often struggles with semi-transparent elements like fine hair, motion blur, and out-of-focus edges, resulting in "green spill" or harsh alpha edges. CorridorKey utilizes a neural network to analyze every pixel in a frame, predicting the true, un-premultiplied foreground color alongside a clean, linear alpha channel. By accepting sRGB or linear images and coarse black-and-white alpha hints, it bridges the gap between rough masks and high-fidelity output. It is engineered specifically for professional VFX workflows, supporting 16-bit and 32-bit EXR files to ensure maximum dynamic range and color integrity for compositors working in pipelines like Nuke or After Effects.

Installation

Installation is managed via the uv package manager for optimal environment isolation. First, ensure you have Python and uv installed. For NVIDIA GPU users, run uv sync --extra cuda. Apple Silicon users should use uv sync --extra mlx, while those on standard hardware can run uv sync for CPU fallback. After syncing dependencies, download the model file from Hugging Face and place it in the CorridorKeyModule/checkpoints/ directory. If you require automatic alpha hint generation, the skill supports integrating GVM or VideoMaMa for advanced segmentation, though these require substantial VRAM resources.

Use Cases

CorridorKey is ideal for independent filmmakers and VFX artists who need to extract high-quality foregrounds from challenging footage. It excels in extracting fine details like stray hair, smoke, or translucent fabric where manual rotoscoping or standard chroma keying fails. It is perfect for fast-tracking post-production on green screen shoots where lighting conditions were less than perfect.

Example Prompts

1. "Open the CorridorKey wizard and guide me through processing the green screen footage located in the /VFX/shots/production_01 folder."
2. "Run the inference on the current sequence using the CUDA device, ensuring the output is saved as 32-bit EXR for my color grading pipeline."
3. "List all available shots in the project folder and start the keying process for the clip named 'studio_interview_04' using the provided alpha mask."

Tips & Limitations

To achieve the best results, use a high-quality, non-compressed source image. While the neural network can reconstruct fine detail from rough masks, the accuracy of the alpha hint significantly influences the stability of the output. Be aware that running deep learning modules like GVM requires significant hardware resources, particularly VRAM. Always ensure your NVIDIA drivers are updated to version 12.8+ to maintain compatibility with the inference engine. If you encounter artifacts, try refining your coarse mask or checking your source footage for excessive compression noise.