[Custom Node] Phase Congruency Edge
https://github.com/bemoregt/ComfyUI_PhaseCongruencyEdge
---
Phase Congruency Edge — A Smarter Edge Detector for ComfyUI
If you've ever used ControlNet with Canny or Depth maps, you know how powerful structural
guidance can be. But Canny has a dirty secret: it breaks under different lighting conditions
and contrast levels. Crank up the exposure on your image and the edge map completely changes.
Phase Congruency Edge solves this problem at the mathematical level.
---
What Makes It Different?
Standard edge detectors like Canny or Sobel work by measuring intensity gradients — how fast
pixel brightness changes. That means their output is directly tied to lighting, exposure, and
local contrast.
Phase Congruency works differently. It looks at the Fourier frequency components of an image
and asks: at which pixels do all these components arrive in phase at the same time? Those
points are the true features — edges, corners, ridges — and they stay stable no matter how
you change the brightness or contrast.
The result is an edge map that is:
- Illumination invariant — same result under any lighting
- Contrast invariant — works on flat, low-contrast images too
- Perceptually accurate — detects what the human eye sees as a boundary
- Multi-scale — captures both fine detail and broad structure simultaneously
---
How to Use It in ComfyUI
Install the node into your custom_nodes folder:
cd ComfyUI/custom_nodes
git clone https://github.com/bemoregt/ComfyUI_PhaseCongruencyEdge.git
Restart ComfyUI and find the node under image/filters → Phase Congruency Edge.
Connect any IMAGE to it and it outputs a clean grayscale edge map — ready to feed into
ControlNet, use as a mask, or composite directly into your workflow.
---
Key Parameters
┌────────────────┬─────────────────────────────────────────────────────────────────────┐
│ Parameter │ What it does │
├────────────────┼─────────────────────────────────────────────────────────────────────┤
│ nscale │ How many frequency scales to analyze (more = richer detail) │
├────────────────┼─────────────────────────────────────────────────────────────────────┤
│ norient │ How many directions to scan (more = smoother, more isotropic edges) │
├────────────────┼─────────────────────────────────────────────────────────────────────┤
│ min_wavelength │ Smallest feature size to detect (lower = finer edges) │
├────────────────┼─────────────────────────────────────────────────────────────────────┤
│ k │ Noise suppression strength (higher = cleaner but less sensitive) │
└────────────────┴─────────────────────────────────────────────────────────────────────┘
The defaults (nscale=4, norient=6, k=2.0) work well for most images right out of the box.
---
Practical Use Cases
- ControlNet preprocessing — a more stable alternative to Canny for stylized or unevenly lit
input images
- Consistent lineart extraction — works reliably across scanned sketches, photos, and renders
- Medical / scientific images — extracts meaningful structure where contrast is naturally low
- Artistic workflows — produces clean, smooth edge lines with a unique character compared to
gradient-based methods
- Masking and compositing — use the edge map to drive inpainting regions or blend layers
---
Background
This algorithm was developed by Peter Kovesi in 1999 and published in the Videre: Journal of
Computer Vision Research. It has been widely used in computer vision research for decades but
has rarely been available as a plug-and-play node in generative AI workflows — until now.
---
Links
- GitHub: https://github.com/bemoregt/ComfyUI_PhaseCongruencyEdge
- Reference: Kovesi, P. (1999). Image Features from Phase Congruency. Videre, 1(3).
---
Give it a try and let me know what workflows you build with it. Feedback and pull requests
are welcome!
---
Tags: ComfyUI, custom node, edge detection, ControlNet, preprocessing,
image processing