AI keying vs Delta Keyer: neural unmixing compared to DaVinci Resolve's built-in chroma keyer
Delta Keyer ships free with every copy of DaVinci Resolve. It lives on the Fusion page. You sample a background color, adjust a few controls, and get a matte. For most green screen work, it does the job well.
AI keying is a different approach entirely. Instead of comparing pixel colors to a sampled screen value, neural unmixing models predict the foreground, background, and alpha channel from the image content itself. The model has learned what edges, hair, glass, and motion blur look like. It does not need a clean, evenly lit screen to produce a good matte.
Both methods are free to use. Both work with DaVinci Resolve. They solve the same problem in fundamentally different ways.
This page compares them honestly. Delta Keyer wins in some situations. AI keying wins in others. Knowing when to reach for each tool will save you hours of frustration.
What is Delta Keyer?
Delta Keyer is the primary chroma keyer built into DaVinci Resolve's Fusion page. It is available in both the free and Studio versions of Resolve. Blackmagic Design included it as the default keying tool, and it has been part of Fusion since the standalone Fusion era, long before Blackmagic acquired it.
Delta Keyer is a chroma-difference keyer. It works by comparing each pixel's color to a sampled background screen color. Pixels that match the screen color become transparent. Pixels that do not match remain opaque. Everything in between gets a partial alpha value.
The tool offers direct controls for:
- ☼ Background color sampling (eyedropper or color picker)
- ☼ Clean plate input for uneven screen lighting
- ☼ Matte controls (threshold, softness, post-erode/dilate)
- ☼ Fringe correction to clean up edge artifacts
- ☼ Spill suppression to remove green or blue color contamination on the subject
Because Delta Keyer is a Fusion node, it sits inside the node tree alongside every other Fusion tool. You can pipe its output into garbage mattes, edge blurs, color corrections, and any other compositing operation without leaving the application.
What is AI keying?
AI keying uses a trained neural network to separate foreground from background. The approach is called neural unmixing. Rather than looking at color values, the model examines the entire image and predicts which pixels belong to the subject and which belong to the screen.
The technology behind CorridorKey (open-sourced by Corridor Digital) trains on thousands of green screen shots with known ground truth mattes. The model learns patterns that color-based keyers cannot detect: the way hair scatters light, how motion blur smears edges, how semi-transparent materials transmit and filter the screen color behind them.
AI keying tools based on CorridorKey typically work as batch processors. You export frames from your timeline, run them through the model, and get back a clean foreground image plus an alpha matte for each frame. These outputs come back into your compositor as standard image sequences.
For a deeper explanation of how neural unmixing works and why it handles difficult footage differently, see our pillar guide to AI chroma keying.
Comparison table
| Delta Keyer | AI keying (CorridorKey-based) | |
|---|---|---|
| Type | Color-difference keyer | Neural unmixing |
| Speed | Real-time in Fusion | Batch processing (seconds per frame) |
| Integration | Native Fusion node | External tool, outputs imported back |
| Hair and fine detail | Requires careful tuning | Strong out of the box |
| Motion blur | Struggles with fast movement | Handles well |
| Spill suppression | Built-in controls | Handled during unmixing |
| Semi-transparency | Limited | Good |
| Cost | Free with Resolve | Free (open source / source-available) |
| Learning curve | Moderate (Fusion knowledge) | Low (batch process, fewer controls) |
| GPU requirements | Any GPU Resolve supports | NVIDIA GPU with sufficient VRAM |
| Clean plate support | Yes (dedicated input) | Not needed |
Where Delta Keyer wins
Delta Keyer has real advantages that AI keying does not replace.
Built into Resolve. There is nothing to install, nothing to export, nothing to import back. You add the node, sample the screen, and start working. The entire round-trip from footage to final composite happens inside one application.
Real-time playback. Delta Keyer processes frames at playback speed on modern hardware. You see results immediately. AI keying requires exporting frames, processing them in a separate tool, and importing the results. That round-trip can take minutes to hours depending on shot length.
Node-based compositing. Delta Keyer's output feeds directly into Fusion's node tree. You can stack it with garbage mattes, edge treatments, light wraps, defocus, grain matching, and color correction. Every adjustment is non-destructive and connected. AI keying outputs are static image sequences that require additional nodes to integrate.
Clean plate support. Delta Keyer accepts a clean plate (an empty frame of just the green screen) as an input. This compensates for uneven lighting, shadows on the screen, and wrinkles in the fabric. It is a simple and effective way to handle imperfect screens.
Decades of VFX industry use. Chroma-difference keying is a well-understood technique. Tutorials, forum answers, and troubleshooting advice are everywhere. If you hit a problem with Delta Keyer, someone has solved it before.
No extra software. Delta Keyer requires nothing beyond DaVinci Resolve itself. No Python environment, no NVIDIA CUDA toolkit, no command-line tools. If Resolve runs on your system, Delta Keyer runs on your system.
Any GPU. Resolve supports NVIDIA, AMD, Intel, and Apple Silicon GPUs. Delta Keyer works on all of them. Most AI keying tools require an NVIDIA GPU with CUDA support and enough VRAM to load the model.
Where AI keying wins
AI keying handles specific problems that color-based keyers fight with.
Hair and fine detail. Individual hair strands scatter light. They are partially transparent and pick up color from the screen behind them. A color-based keyer sees these pixels as "mostly green" and makes them transparent, erasing the hair. Neural unmixing recognizes hair as a structural element and preserves it, even when the color data is ambiguous.
Motion blur. When a subject moves quickly, the edges of their body blur into the background. These blurred pixels are a mix of subject and screen colors. Delta Keyer treats them as partial transparency, which creates hard edges on blurred subjects. AI keying models understand motion blur as a physical phenomenon and produce soft, natural edges that match the original motion.
Semi-transparent materials. Veils, glass, smoke, and sheer fabrics transmit light from the screen. Color-based keyers remove these areas because they match the screen color. Neural unmixing can distinguish between "this is the screen" and "this is a transparent object in front of the screen."
Less manual tweaking. A difficult shot in Delta Keyer might require 20 minutes of threshold adjustments, edge refinement, garbage masking, and spill correction. The same shot through an AI keyer often produces a usable matte with no manual adjustment. The model has already learned what a good matte looks like.
Consistent results across lighting variation. If the green screen lighting changes within a shot (a cloud passes, a practical light shifts), Delta Keyer's sampled color becomes less accurate over time. AI keying evaluates each frame independently and adapts to changing conditions without manual keyframing.
When to use which
The choice between Delta Keyer and AI keying usually comes down to the footage quality and the complexity of the subject.
Use Delta Keyer when:
- ☼ Your green screen is evenly lit and clean
- ☼ The subject has hard, well-defined edges (solid costumes, props, simple hair)
- ☼ You need real-time feedback while compositing
- ☼ The shot involves heavy Fusion compositing that benefits from staying in the node tree
- ☼ You are working on a deadline and the footage cooperates
Use AI keying when:
- ☼ Hair, fur, or fine wispy detail is prominent in the shot
- ☼ The footage has motion blur on edges
- ☼ Semi-transparent materials are in frame (veils, glass, smoke)
- ☼ The green screen has uneven lighting, wrinkles, or shadows you cannot fix
- ☼ You have tried Delta Keyer and spent more than 15 minutes fighting edge quality
The practical reality
Most DaVinci Resolve editors will use Delta Keyer first. It is right there in Fusion. It handles the majority of green screen shots without problems. AI keying becomes the tool you reach for when Delta Keyer is not giving you clean results, particularly on problem shots with complex edges or poor screen conditions.
A common workflow on a project with many green screen shots: key 80% of them with Delta Keyer, flag the difficult ones, and send those through an AI keyer for better results.
Workflow: using AI keying with DaVinci Resolve
Here is how to combine an AI keying tool like EZ-CorridorKey with DaVinci Resolve.
Step 1: Export frames from Resolve
On the Edit or Cut page, set in and out points on your green screen clip. Go to File > Export > Individual Clips, or render the clip from the Deliver page as an image sequence (PNG or EXR). Use a dedicated folder for the export.
Step 2: Process with the AI keyer
Run the exported frames through the CorridorKey-based tool. EZ-CorridorKey processes each frame and outputs two image sequences:
- ☼ A clean foreground (the subject with the green screen removed, replaced with black or transparency)
- ☼ An alpha matte (white where the subject is, black where the screen was)
Processing speed depends on your GPU. Expect 1 to 5 seconds per frame on a mid-range NVIDIA card.
Step 3: Import results into Fusion
On the Fusion page, add two MediaIn nodes: one for the clean foreground sequence and one for the alpha matte. Use a Channel Booleans or Matte Control node to combine the foreground RGB with the alpha channel. This gives you a properly keyed element inside Fusion's node tree.
From here, you have the same compositing flexibility as a Delta Keyer output. Add light wraps, edge treatments, color matching, defocus, and grain. The matte is just another channel in the Fusion pipeline.
Step 4: Composite normally
Merge the keyed foreground over your background plate using a Merge node. The compositing workflow from this point is identical to working with any other keying method in Fusion.
Frequently asked questions
Can I use Delta Keyer and AI keying on the same project?
Yes. There is no conflict between the two approaches. Use Delta Keyer for shots where it produces clean results and switch to AI keying for the difficult shots. Both methods output the same thing: a foreground element with an alpha channel. Fusion does not care how the matte was generated.
Does AI keying work with blue screen footage?
Yes. EZ-CorridorKey v2.0 added full blue screen support alongside green screen keying. The original CorridorKey was trained primarily on green screen footage, so other forks may vary. Delta Keyer works equally well with green and blue screens because it samples whatever background color you give it.
Is EZ-CorridorKey ready for production use?
EZ-CorridorKey is a source-available work in progress with over 4,000 stars on GitHub. It is the #1 fork of Corridor Digital's CorridorKey project and is actively developed by EZSCAPE. Version 2.0 added blue screen support alongside the original green screen keying. It produces good results on many shots, but it is not a finished commercial product. Test it on your footage before committing to it for a deadline project. Available on Gumroad.
Do I need DaVinci Resolve Studio for either method?
No. Delta Keyer is available in the free version of DaVinci Resolve. AI keying tools like EZ-CorridorKey are also free. You can build a complete green screen compositing workflow without spending anything on software. The only hardware requirement for AI keying is an NVIDIA GPU with enough VRAM to run the model.
EZ-CorridorKey is developed by Ed Zisk at EZSCAPE, creator of DITHERON, CRTified, and Cucolori.