OpenCV integral2_1

The pattern should be familiar by now: OpenCV integral2_1 and OpenCV integral2_0 are the same node. This pack auto-generates one node per overload found in OpenCV's type stubs, cv2.integral2 is listed twice (MatLike and UMat variants), and the _1 here is that duplicate. Same inputs, same two outputs, no behavioral difference. Read the integral2_0 article for the full mechanics; this page covers the practical side and the numbering trap.

What it does

Computes two summed-area tables in a single pass: sum, the ordinary integral image, and sqsum, the integral of squared pixel values. Together they let you compute per-rectangle mean and variance in constant time - variance being mean(squares) − mean². That makes it the right tool for local contrast normalization, per-tile statistics, and texture/region analysis. It's a building block, not an end-user node: you'll rarely drop it into a standard txt2img graph, and when you want the results of such analysis you usually reach for adaptiveThreshold_0 or blur_0 in the same pack.

Inputs and outputs

• src - NPARRAY, your image via Image2Nparray.
• sdepth - INT, depth for the sum table (4 = CV_32S for 8-bit sources).
• sqdepth - INT, depth for the squared-sum table - use 5 (CV_32F) or 6 (CV_64F), never integer, or you overflow.
• sum, sqsum - optional out-parameters, safe to ignore.
• Outputs nparray_0 (sum), nparray_1 (squared sum).

Installing

Standard for the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/geroldmeisinger/opencv-comfyui
cd opencv-comfyui
pip install -r requirements.txt

or ComfyUI Manager → "opencv-comfyui". No model downloads; requirements.txt is opencv-contrib-python, numpy, torch.

Gotchas

Two things to remember, both from the pack's nature rather than this specific node. First, the two outputs are tables, not images - previewing them via Nparrays2Image gives clipped nonsense because the values run far beyond 255 and nothing rescales them. Second, integral2_1 is not the "advanced" version. The _1 suffix is pure overload numbering; there is nothing this node does that integral2_0 doesn't. Use whichever a workflow references, and keep batch_size == 1 in all cases.