bemoregt/ComfyUI_NotchFilter
NotchFilter custom node for comfyui
ComfyUI Spectrum Notch Filter
A ComfyUI custom node for removing periodic noise from images via frequency-domain notch filtering.
Takes an FFT amplitude spectrum image as input, masks the noise peaks, and outputs the cleaned spectrum ready for inverse FFT.
Nodes
Spectrum Notch Filter (Auto)
Automatically detects periodic noise peaks in a spectrum image using local maxima analysis and applies circular notch masks.
Inputs
| Name | Type | Default | Description |
|---|---|---|---|
spectrum |
IMAGE | — | FFT amplitude spectrum image |
threshold_rel |
FLOAT | 0.5 | Peak detection threshold relative to the maximum value. Lower → more peaks detected |
min_distance |
INT | 10 | Minimum distance (px) between detected peaks. Prevents double-detection of the same peak |
notch_radius |
INT | 8 | Radius (px) of the circular mask applied around each peak |
protect_dc |
INT | 20 | Radius (px) around the DC component (image center) to ignore during detection |
feather |
INT | 2 | Gaussian sigma for soft mask edges. 0 = hard binary mask |
Outputs
| Name | Type | Description |
|---|---|---|
filtered_spectrum |
IMAGE | Spectrum with noise peaks zeroed out |
notch_mask |
IMAGE | White-on-black mask image (white = removed region) |
preview |
IMAGE | Input spectrum annotated with yellow rings at detected peaks and a blue ring at the DC zone |
peak_positions |
STRING | JSON array of detected peaks — wire this into SpectrumNotchManual for manual refinement |
Spectrum Notch Filter (Manual)
Applies user-drawn circular notch masks to a spectrum image. Features an interactive canvas widget embedded directly in the node.
Inputs
| Name | Type | Default | Description |
|---|---|---|---|
spectrum |
IMAGE | — | FFT amplitude spectrum image |
notch_points |
STRING | [] |
JSON array of notch circles. Automatically updated by the canvas widget |
feather |
INT | 2 | Gaussian sigma for soft mask edges |
protect_dc |
INT | 0 | Optional DC protection radius (applied in addition to drawn circles) |
Outputs
| Name | Type | Description |
|---|---|---|
filtered_spectrum |
IMAGE | Spectrum with selected peaks masked |
notch_mask |
IMAGE | Mask image |
Canvas Widget Controls
After running the node once, the input spectrum is displayed as the canvas background.
| Action | Effect |
|---|---|
| Left-click | Add a notch circle at the clicked position |
| Right-click | Remove the circle under the cursor |
| Shift + drag | Resize the nearest circle in real time |
| Radius slider | Set the default radius for newly added circles |
| Symmetry checkbox | Automatically add the conjugate-symmetric counterpart of each circle (recommended for standard FFT spectra) |
| Clear button | Remove all notch circles |
The blue crosshair marks the DC component (image center). Circle indices are shown inside each circle.
Recommended Workflow
[Image → Spectrum node]
│
▼
[SpectrumNotchAuto] ──── preview ──────► [PreviewImage] (inspect detected peaks)
│
│ peak_positions (STRING)
▼
[SpectrumNotchManual] ◄── run once, then click to refine on canvas
│
│ filtered_spectrum
▼
[Inverse FFT node]
│
▼
[Output Image]
- Connect your amplitude spectrum to SpectrumNotchAuto.
- Tune
threshold_relandnotch_radiusuntil thepreviewoutput looks correct. - Wire
peak_positionsintonotch_pointsof SpectrumNotchManual. - Run the Manual node — the spectrum appears on the canvas.
- Click to add or remove individual notch circles, then re-run.
Installation
# From your ComfyUI custom_nodes directory
git clone https://github.com/your-username/ComfyUI_NotchFilter
cd ComfyUI_NotchFilter
pip install -r requirements.txtRestart ComfyUI. The two nodes appear under image/frequency in the node menu.
Dependencies
numpyscipyPillowtorch(already provided by ComfyUI)
How It Works
Peak Detection (Auto node)
spectrum image → grayscale → maximum_filter (size = 2·min_distance + 1)
→ local maxima where value ≥ threshold_rel · max
→ exclude DC zone
→ circular masks around each peak (+ Gaussian feather)
→ filtered = spectrum × (1 − mask)
Uses scipy.ndimage.maximum_filter — no additional dependencies beyond scipy.
Notch peak_positions JSON Format
[
{"x": 128, "y": 64, "r": 8},
{"x": 128, "y": 192, "r": 8}
]x,y: pixel coordinates in the spectrum image (origin = top-left)r: notch circle radius in pixels
You can also type or paste this JSON directly into the notch_points field.
Example: Removing Grid Artifacts
Grid-pattern noise in an image produces a regular array of bright dots in the FFT spectrum (excluding the DC center). Set protect_dc to ~20, lower threshold_rel until all the artifact dots are highlighted in the preview, then run the filter. Wire the result to your inverse FFT node to recover the cleaned image.