nikopueringer/CorridorKey ? reverse-engineered prompt
Reverse engineered prompt
I want to build a better green screen keying tool in Python. I'm tired of the standard keyers that struggle with fine hair, motion blur, and transparent edges. This tool should use an AI model to perfectly separate the subject from the green background.
Instead of just creating a harsh matte, it should intelligently figure out the true colors of those tricky semi-transparent pixels, essentially reconstructing the foreground image as if the green screen was never there. It should output a clean foreground layer and a perfect alpha channel.
The tool needs to handle high-res footage like 4K and export to a professional format like 16-bit EXR sequences so I can use the results in Nuke or Resolve. It should have a simple command-line interface and maybe even some drag-and-drop scripts for Mac and Windows. It should use the GPU for processing—make sure it works with NVIDIA CUDA and Apple Silicon. For setup, let's use uv to manage the Python environment and dependencies, and it should automatically download the main AI model from HuggingFace on first run.
Are you gonna build this?
make sure you review the code using coderabbit