OpenCV erode_1

Straight answer: erode_1 is a duplicate of erode_0. Same inputs, same output, same generated code - if you open the pack source and diff the two classes, they're copy-paste twins. You can use either one and get identical results. If you're deciding, just click erode_0 and move on with your life.

Why does the twin exist? opencv-comfyui (geroldmeisinger/opencv-comfyui) is auto-generated: it parses OpenCV's type stubs and emits one node per function overload, numbered _0, _1, etc. OpenCV's erode is declared a couple of times in the stubs (mostly differing in how the optional dst out-parameter shows up), and when the overloads are functionally identical, the generator spits out two lookalike nodes. It's a byproduct of the pipeline, not a feature you're supposed to choose between.

What it does

Morphological erosion: erases small white specks, shrinks bright regions, and thins features in a binary image or mask. In a typical ComfyUI workflow that means cleaning up a mask - killing the 1-pixel noise and separating blobs that are just barely touching - before you send it to something downstream. It's the counterpart to dilate (which grows bright regions), and together they're the standard mask-tidy pair. Deterministic, instant, no model involved.

Inputs and outputs

• src (NPARRAY) - the image or mask to erode.
• kernel (NPARRAY) - the structuring element as a matrix, typically built with the pack's getStructuringElement_0 node.
• anchor (STRING) - literal string, default (-1, -1) for center.
• iterations (INT) - how many passes; bump this before building a giant kernel.
• borderType (INT) - border enum, 0 (BORDER_CONSTANT) is the safe default.
• borderValue (STRING) - literal border value, e.g. 0.
• dst (NPARRAY, optional) - OpenCV out-parameter, leave unconnected.
• Output: nparray - the eroded result.

Wiring it up

Same pattern as every node in this pack: Image2Nparray in, Nparrays2Image out, and remember this pack speaks BGR. The kernel is the one input you must actually build - there's no "size" widget, you feed a structuring element nparray from getStructuringElement_0.

Install and gotchas

Install via ComfyUI Manager (search "opencv-comfyui") or git clone https://github.com/geroldmeisinger/opencv-comfyui into ComfyUI/custom_nodes, then restart. Requirements are opencv-contrib-python, numpy, torch; no models to fetch. The pack's own warning: "Expect dragons!" - the string fields parse as Python literals, so invalid syntax (<unknown>, line 0) means you typed anchor or borderValue wrong. That and the -215 grayscale assertion (fix with cvtColor code 6) are the two errors you'll actually meet. Full walkthrough and troubleshooting live on the erode_0 page - this one is just the twin.