CorridorKey 2: Background Removal Without Chroma Key Using AI

The Corridor Crew team has released an updated version of the tool for removing green and blue backgrounds. The new version works even in scenes without a chroma key, predicting a mask and clean foreground for each frame.

Support for Complex Objects

The tool combines predictions from three neural networks: C-RADIOv4 by NVIDIA, MoGe-2 by Microsoft, and RVM by Google DeepMind. This improves accuracy when processing hair, fur, semi-transparent objects, and out-of-focus areas.

Users can choose one of three hint options to isolate the object. The No Hint model automatically determines boundaries, First-Frame Hint uses the first frame, and Full Hint Video applies hints for every frame.

Technical Specifications and Requirements

CorridorKey 2 operates at the video's original resolution with temporal stabilization via Hanning windows, eliminating flickering along the mask edges. It supports HDR and exports frame sequences in OpenEXR format with linear color space.

An RTX 30-series or newer graphics card is required. Minimum memory requirements are 4 GB VRAM for 2K resolution and 6 GB VRAM for 4K. The software is available for Windows and Linux.

Plugins are available for DaVinci Resolve, Nuke, After Effects, Premiere Pro, and Blender on Windows and Mac. A standalone application is also available. The code is open source on GitHub, and model weights are hosted on Hugging Face.