OpenCV selectROI_3

selectROI_3 is selectROI_2's UMat overload - the same no-window-name cv2.selectROI call, the same inputs, the same behavior, and no difference you'll ever observe from inside ComfyUI. The pack generated a numbered node for every OpenCV type-stub overload, and this is the second copy of the "skip the window name" variant.

What it does

What it does: pop an interactive window over your image, let you drag a rectangle, and return the region. Inputs: img (NPARRAY), plus showCrosshair, fromCenter, and printNotice booleans. Output is a STRING-labeled literal that in practice carries the raw (x, y, w, h) tuple - one of this pack's genuine dragons, since it isn't serialized to a text string. It blocks the workflow until you click, needs a real display (headless servers error), and only takes batch_size==1 images. The full context lives in the selectROI_0 and selectROI_2 articles; nothing here changes for this variant.

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/geroldmeisinger/opencv-comfyui
pip install opencv-contrib-python

Install: ComfyUI Manager → search "opencv-comfyui", or the commands above, then pip install opencv-contrib-python for the dependency. No model downloads.

Verdict

The complete family map, once and for all: selectROI_0/1 are the named-window overloads (MatLike/UMat), selectROI_2/3 are the default-window ones (MatLike/UMat). Within each pair the variants are interchangeable - if a workflow references one and you have the other, it runs as-is. This is a short article because it's the fourth copy of one function.

The interaction rules apply to all of them: drag to select, Enter/space to confirm, c to cancel, and the graph stays blocked until you act. And if your ComfyUI runs headless - a Docker container or a remote box - none of the selectROI nodes will work, because there's no window to draw in.