HT Moiré Removal

Photograph a computer monitor or scan a printed magazine and you know the enemy: those wavy, color-shifting interference bands that shimmer across fine patterns. That's moiré, and it wrecks perfectly good images right before they enter your pipeline. HT Moiré Removal is the pack's dedicated de-moiré filter - a frequency-domain clean-up that targets the interference pattern while leaving the underlying image mostly alone. It's the node you reach for when your source is a scan or a screen photo and you want a clean base for further processing.

How it works

It stacks three classic filters, each toggleable:

• Median filter (median_enabled, kernel default 7) - a spatial pass that kills small periodic noise. Cheap, first line of defense.
• Butterworth filter (butterworth_enabled) - the frequency-domain workhorse. The image goes into the Fourier domain, a Butterworth low-pass attenuates the high-frequency band where moiré lives, and it comes back out. The butterworth_a/b (0.75/1.25) shape the filter's roll-off, butterworth_cutoff (30) sets where it starts cutting, and butterworth_order (2) controls how aggressive the cut is.
• Notch filter (notch_enabled, off by default) - the surgical option. Screen-photo moiré isn't random; it concentrates at specific frequencies set by the pixel grid and camera angle. Notch filtering removes a band of frequencies rather than everything above a cutoff. The notch_h_spacing/notch_v_spacing (135/200) and notch_size (15) define where those bands sit. Turn this on when the Butterworth pass leaves periodic residual bands - it's the difference between "softened the whole image" and "removed the actual pattern."

Everything runs tiled for memory - tile_size_mode (1024/768/512/Conservative/Aggressive), tile_overlap (128), edge_padding (64) - with global_normalization on to keep brightness consistent across tiles, which is exactly how you avoid the seams tiled processing usually introduces. resolution_scaling_factor (default 1) lets you pre-scale spatial parameters for very large inputs. Output: processed_image, straight into a preview or the next stage.

When to use it (and when not to)

Use it on scanned illustrations, photographed screens, digitized print - sources with genuine periodic interference. Don't expect it to "fix" a jpeg-compression mosaic or sensor noise; that's a different problem (the pack's other filters, like the denoiser, are aimed there). And remember frequency filtering trades detail for clean: start with median on, Butterworth on, notch off, then enable notch only if the periodic bands survive.

Installing

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/ArtHommage/HommageTools.git
cd HommageTools && pip install -r requirements.txt

restart, or Manager → "HommageTools for ComfyUI". It uses OpenCV/numpy for the median and FFT passes, which a standard ComfyUI environment already has - no extra model files. Given the pack's "perpetually in alpha" disclaimer, re-check results after updates: frequency-domain parameter behavior is exactly the kind of thing that gets tuned between versions.