OpenCV goodFeaturesToTrackWithQuality_1

goodFeaturesToTrackWithQuality_1 is goodFeaturesToTrackWithQuality_0 wearing a different name tag. Same required inputs (image, maxCorners, qualityLevel, minDistance, mask, blockSize, gradientSize, useHarrisDetector, k), same optional out-params (corners, cornersQuality), same two outputs (nparray_0 corners, nparray_1 quality scores). Feed both the same image and you get byte-identical results. This is the pack's duplicate-of-a-duplicate problem: the auto-generator parsed OpenCV's overloads, wrapped each one, and two of the goodFeaturesToTrackWithQuality overloads collapsed into the same node.

What you're actually looking at

The underlying function is the same Shi-Tomasi corner detector with per-corner quality scores returned, the newer 4.5.3+ API. The workflow value is the second output: the N×1 float32 quality array lets you filter corners by their individual strength after the fact, instead of trusting only the global qualityLevel. For tracking, alignment, or region matching that's a real tool. All of that is covered properly in the _0 article - this node adds nothing to the story.

Which one to use

Use goodFeaturesToTrackWithQuality_0. There is no technical reason, but there is a practical one: _0 is the canonical first overload, so shared workflows and other people's graphs will tend to reference it. Using _1 invites a moment of "wait, is this different?" every time you or someone else loads the graph. The duplicates in this pack - goodFeaturesToTrack_0/_1, _2/_3, and this pair - are all the same species: completeness-over-curation artifacts the author warned about with "Expect dragons!" in the README.

Install

Same pack, same steps:

cd ComfyUI/custom_nodes
git clone https://github.com/geroldmeisinger/opencv-comfyui

or via ComfyUI Manager (search "opencv-comfyui"), restart. The pack depends on opencv-contrib-python; numpy and torch already ship with ComfyUI. No models, no API keys. Known startup clash to remember: Cannot import name 'guidedFilter' from 'cv2.ximgproc' means you have multiple OpenCV packages installed - uninstall duplicates and keep one wheel.

The honest bottom line

Nothing here is wrong with your install or your understanding. The twin nodes are the pack's own brand of clutter. If you need quality-scored corners, goodFeaturesToTrackWithQuality_0 does the job, and this one is the spare. Treat _1 as documentation of how auto-generated packs work, delete it from your graph, and move on - there's no hidden setting, no second algorithm, no better result waiting behind the _1 label.