Repolode
Submit
nikopueringer

nikopueringer/CorridorKey2.0

AI green and blue screen keying model that produces alpha mattes and foreground color from video, run from Python with an NVIDIA GPU.

Open-source alternative to:KeylightUltimatte

View on GitHubWebsite
Stars
886
Forks
98
Open issues
5
Last commit
7 days ago

What is it?

CorridorKey 2.0 is a video keying model for green and blue screen footage. It predicts an alpha matte and a clean foreground color for each frame, and can run with no hint, with a single first-frame matte hint, or with a full hint video for every frame.

The repository provides the Python inference pipeline with a command-line tool (infer.py) and a guided drag-and-drop launcher for Windows and Linux. Outputs include matte previews, foreground previews, a checkerboard review render, and a cutout with despill control. Output can be written as ffmpeg-encoded video or as OpenEXR image sequences with unclipped HDR values, and inputs can be video files or frame directories including linear EXR.

The code requires an NVIDIA GPU with CUDA-enabled PyTorch and ffmpeg on the PATH. Memory use is managed through Hann window sizes and a low-VRAM mode for long clips or smaller cards. The model and code are released under a non-commercial-style license with extra terms, so check the license before commercial use.

Key features

  • 01Produces alpha mattes and foreground color from green or blue screen video.
  • 02Accepts no hint, a first-frame matte hint, or a full hint video per frame.
  • 03Writes ffmpeg video previews or OpenEXR sequences with unclipped HDR values.
  • 04Includes checkerboard review and cutout outputs with adjustable despill.
  • 05Automatically picks temporal window size and low-VRAM mode based on GPU memory.

Use cases

  • Keying green or blue screen shots in VFX and compositing pipelines.
  • Generating mattes for review renders before final compositing.
  • Processing linear EXR plates for HDR compositing workflows.
  • Running batch keying on a folder of clips from the command line.

Quick start

  1. 1

    Create a Python virtual environment

    python -m venv venv
  2. 2

    Activate the virtual environment

    source venv/bin/activate
  3. 3

    Install CUDA PyTorch and the remaining dependencies

    pip install torch==2.9.1 torchvision==0.24.1 --index-url https://download.pytorch.org/whl/cu128 && pip install -r requirements.txt
  4. 4

    Download the model weights (about 5 GB) and run inference on a clip

    python scripts/download_weights.py

Things to know

  • Requires an NVIDIA GPU with at least 4 GB VRAM (RTX 30-series or newer recommended).
  • Needs ffmpeg on your PATH and a CUDA-enabled PyTorch setup.
  • Model weights are about 5 GB to download.

In collections

Computer vision libraries & tools

Topics

#green-screen#chroma-key#video-matting#compositing#pytorch#computer-vision#blue-screen#keying#machine-learning#vfx

FAQ

What is CorridorKey2.0?
AI green and blue screen keying model that produces alpha mattes and foreground color from video, run from Python with an NVIDIA GPU. CorridorKey 2.0 is a video keying model for green and blue screen footage. It predicts an alpha matte and a clean foreground color for each frame, and can run with no hint, with a single first-frame matte hint, or with a full hint video for every frame.
Is CorridorKey2.0 free to use? What license does it have?
CorridorKey2.0 is open source under the Other license. Check the license terms before using it in commercial products.
Is CorridorKey2.0 still maintained?
Actively maintained. The last commit to CorridorKey2.0 was on Oct 15, 2026.
What can I use CorridorKey2.0 for?
Keying green or blue screen shots in VFX and compositing pipelines. Generating mattes for review renders before final compositing. Processing linear EXR plates for HDR compositing workflows. Running batch keying on a folder of clips from the command line.
How do I install CorridorKey2.0?
Create a Python virtual environment: python -m venv venv Activate the virtual environment: source venv/bin/activate Install CUDA PyTorch and the remaining dependencies: pip install torch==2.9.1 torchvision==0.24.1 --index-url https://download.pytorch.org/whl/cu128 && pip install -r requirements.txt Download the model weights (about 5 GB) and run inference on a clip: python scripts/download_weights.py
Where can I get CorridorKey2.0?
The source code and installation instructions are on GitHub: https://github.com/nikopueringer/CorridorKey2.0

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects