EZSCAPE

CorridorKey workflow for DaVinci Resolve: AI keying in your Resolve pipeline

DaVinci Resolve ships with Delta Keyer. For most green screen shots, Delta Keyer is all you need. It handles clean footage with even lighting and hard edges without breaking a sweat.

But some shots fight back. Flyaway hair against a wrinkled screen. Motion blur from a fast-moving arm. Semi-transparent fabric. Uneven lighting that no amount of clip-black adjustment can fix. These are the shots where traditional keyers hit their limits and where CorridorKey picks up the slack.

This guide walks through a practical workflow for using CorridorKey-based tools alongside DaVinci Resolve. The approach is complementary. You keep Delta Keyer as your primary keyer and bring CorridorKey in only for the shots that need it.

CorridorKey implementations for Resolve users

There are three ways to get CorridorKey working with your Resolve pipeline. Each has different tradeoffs.

OFX plugins

Several community developers have packaged CorridorKey's neural unmixing into OFX plugins. These run directly inside Fusion's node graph. You add the node, connect your source, and process without leaving Resolve. The main advantage is zero round-tripping. The main limitation is that OFX wrappers vary in quality and may lag behind the core CorridorKey model updates.

Standalone GUI tools

Desktop applications that wrap the CorridorKey pipeline with a graphical interface. You export frames from Resolve, process them in the standalone tool, and import the results back.

EZ-CorridorKey is the most-starred fork on GitHub with over 4,000 stars. Version 2.0 added blue screen support alongside green. It includes 6 alpha generators (BiRefNet, GVM, VideoMaMa, MatAnyone2, Apple Vision MLX, and manual chroma key), batch processing, and VRAM optimizations that bring the GPU requirement down from the original 24GB. It is source-available and a work in progress. You can download it from Gumroad.

Other standalone forks exist on GitHub. Some optimize for speed, others for quality, others for specific hardware. Search "CorridorKey" and sort by stars or recent activity.

Original CLI

The original Python script from Corridor Digital. Command-line only. Requires 24GB+ VRAM, Python, and CUDA setup. Powerful but not practical for most editors.

Step-by-step standalone workflow

This workflow covers the most common path: using a standalone CorridorKey tool alongside Resolve. The steps apply to EZ-CorridorKey, other GUI forks, and even the CLI with minor adjustments.

Step 1: Identify your problem shots

Play through your timeline and flag the shots where Delta Keyer is struggling. Look for:

☼ Hair edges that show green fringing or lose fine strands ☼ Motion blur at the subject's edges that creates a halo ☼ Semi-transparent materials that Delta Keyer either removes entirely or leaves contaminated with green ☼ Uneven green screen lighting that forces you to pull multiple keys and combine them

If Delta Keyer handles a shot cleanly in a few minutes of adjustment, leave it. CorridorKey processing takes time. Reserve it for the shots that actually need it.

Step 2: Export frames from Resolve

Switch to the Deliver page. This is the fastest way to export frame sequences from Resolve.

☼ Set your render range to the specific clip or In/Out range of the problem shot ☼ Choose "Individual Clips" if you are exporting multiple shots ☼ Set the format to PNG (8-bit, fast processing) or EXR (16-bit/32-bit, maximum quality and color depth) ☼ Match your timeline resolution. Do not scale during export ☼ Match your timeline frame rate ☼ Create a dedicated output folder for each shot, for example: CorridorKey_Exports/Shot_012/

EXR preserves more color data and gives the neural network more information to work with. The quality difference matters most on shots with subtle transparency or fine color gradients. PNG is fine for straightforward shots where speed matters more than marginal quality gains.

Add the job to the render queue and render. You now have a numbered image sequence ready for processing.

Step 3: Process in your CorridorKey tool

Open your CorridorKey application and load the exported frame sequence.

Select an alpha generator. This creates the coarse mask that tells the neural network where your subject is. Different generators work better on different footage. If your tool offers multiple options, try two or three on a single frame and compare before batch processing the entire shot ☼ Set the screen color. Green screen or blue screen, depending on your footage ☼ Set the output format. Match what you exported. If you exported EXR, output EXR. If you exported PNG, output PNG ☼ Run a test on 3-5 frames first. Check the output alpha and foreground color before committing to the full batch. Look for edge artifacts, missed areas, and color contamination ☼ Process the full sequence. Depending on your GPU and frame count, this takes anywhere from a few minutes to an hour or more

The output is two image sequences: a clean foreground (with the green screen removed and replaced with black or transparency) and an alpha matte (a grayscale image where white is fully opaque and black is fully transparent).

Step 4: Import results into Fusion

Switch to Resolve's Fusion page. You need to bring in two elements:

Foreground: The clean foreground image sequence from CorridorKey. Add it as a MediaIn node (or Loader node in Fusion Standalone). Point it to the first frame of the foreground sequence ☼ Alpha matte: The alpha output from CorridorKey. Add it as a second MediaIn/Loader node

If your CorridorKey tool outputs a single RGBA image sequence (foreground with embedded alpha), you only need one MediaIn node. Check your tool's output settings.

Step 5: Combine foreground and alpha

Connect the foreground and alpha using one of these methods.

Method A: Channel Booleans node

☼ Add a Channel Booleans node after your foreground MediaIn ☼ Connect the foreground to the Channel Booleans "Background" input ☼ Connect the alpha matte to the "Foreground" input ☼ Set the operation to copy the luminance (or red channel) of the alpha matte into the alpha channel of the foreground ☼ This gives you an RGBA foreground with the CorridorKey alpha

Method B: Matte Control node

☼ Add a Matte Control node after your foreground MediaIn ☼ Connect the alpha matte to the Matte Control's "Garbage Matte" or "Effect Mask" input ☼ Adjust the matte settings to use the external matte as your alpha source

Both methods produce the same result: a foreground image with a proper alpha channel derived from the CorridorKey output.

Step 6: Merge over your background

☼ Add a Merge node ☼ Connect your alpha-combined foreground to the Merge "Foreground" input ☼ Connect your background plate (or replacement background) to the "Background" input ☼ The composite should look clean at this point. If edges look rough, proceed to Step 7

Step 7: Finishing touches

A raw composite almost always benefits from edge treatment and color matching.

Light wrap. Add a Light Wrap node (built into Fusion) between the Merge and the output. This bleeds background light onto the foreground edges, simulating how real light wraps around a subject. Subtle settings work best. Start with a wrap size of 2-4 pixels and low strength ☼ Edge treatment. Use an Erode/Dilate node to shrink or expand the matte by a pixel if you see a thin dark or bright line at the edges. An Erode value of -0.5 to -1.0 typically cleans up edge artifacts ☼ Color match. The foreground was lit by your green screen studio. The background has its own color temperature and contrast. Use a Color Corrector node on the foreground (before the Merge) to match white balance, exposure, and contrast to the background plate ☼ Spill suppression. CorridorKey's neural unmixing handles most green spill during processing. If you still see green contamination on skin tones or edges, add a Spill Suppression node. Keep adjustments minimal since over-suppression creates unnatural color shifts

Tips for Resolve-specific workflow

Use the Deliver page for exports. The Deliver page is faster than dragging individual clips out or using Fusion's file saver for initial exports. It handles frame numbering, format conversion, and range selection in one step ☼ Name your sequences clearly. Use a naming convention like Shot012_FG_00001.exr and Shot012_Alpha_00001.exr. When you have 15 problem shots, clear naming prevents confusion ☼ Keep frame numbering consistent. If Resolve exports frames starting at 1001 (a common convention), make sure your CorridorKey output also starts at 1001. Mismatched frame numbers cause sync issues in Fusion ☼ Use EXR when quality matters. EXR at 16-bit float preserves highlight and shadow detail that 8-bit PNG clips. For hero shots and close-ups, the extra file size is worth it ☼ Use PNG when speed matters. PNG files are smaller and faster to read/write. For wide shots or footage that will be scaled down, PNG is sufficient ☼ Save Fusion compositions as templates. Once you build your Channel Booleans, Merge, Light Wrap, Edge Erode chain once, save it as a macro or template. Apply it to every CorridorKey shot with just a path change on the MediaIn nodes

When to use Delta Keyer vs CorridorKey

This decision tree covers 90% of keying situations.

Start with Delta Keyer. Pull your key. Adjust matte settings. Check your edges.

Edges look clean, matte is solid? You are done. Stay with Delta Keyer. ☼ Hair is fringing or losing fine strands? Try CorridorKey on this shot. ☼ Motion blur creating halos? Try CorridorKey on this shot. ☼ Transparent materials going opaque or getting destroyed? Try CorridorKey on this shot. ☼ Uneven screen, but hard edges only? Try adjusting Delta Keyer's pre-matte range first. If you can fix it in 10 minutes, stay with Delta Keyer. If not, export for CorridorKey. ☼ Uneven screen plus fine detail? Go straight to CorridorKey. This combination is where traditional keyers struggle most.

The goal is not to process every shot through CorridorKey. The goal is to process only the shots that need it, and get those shots to a quality level that Delta Keyer could not reach.

FAQ

Can I use CorridorKey directly inside the Fusion page?

Only if you use an OFX plugin version. The original CorridorKey and most standalone forks (including EZ-CorridorKey) run outside of Resolve. You export frames, process them externally, and import the results back into Fusion. OFX plugin implementations do run inside Fusion's node graph, but they are community-maintained and may not support the latest model updates.

What frame format should I export from Resolve for CorridorKey?

PNG or EXR. Use PNG (8-bit) for general work where speed and smaller file sizes are priorities. Use EXR (16-bit float or 32-bit float) for hero shots, close-ups, or footage with subtle transparency where maximum color precision matters. Avoid JPEG. The lossy compression introduces artifacts that degrade keying quality. TIFF works but offers no advantage over PNG for 8-bit or EXR for higher bit depths.

Does CorridorKey handle blue screen footage?

The original CorridorKey was designed for green screen. Some forks have added blue screen support. EZ-CorridorKey added blue screen keying in version 2.0. Check your specific CorridorKey tool's documentation. If blue screen is not supported, you can sometimes work around this by hue-shifting your blue screen footage to green, processing, and then color-correcting the foreground back. This is a hack and produces variable results.

How long does CorridorKey processing take per frame?

Processing time depends on resolution, GPU, and the specific implementation. On a modern GPU with 12GB+ VRAM, expect 2-5 seconds per frame at 1080p. 4K frames take longer, roughly 8-15 seconds each. A 10-second shot at 24fps (240 frames) takes approximately 8-20 minutes at 1080p. Plan accordingly and batch process overnight if you have many shots.


CorridorKey and EZ-CorridorKey are developed by Ed Zisk at EZSCAPE, creator of DITHERON, CRTified, and Cucolori. Browse more guides at ezscape.space/learn.