OpenCV CV_16SC_0

CV_16SC_0 computes cv2.CV_16SC(channels) - the OpenCV type constant for a signed 16-bit integer array with N channels. One channels INT in, one int constant out. Another auto-generated calculator node from the opencv-comfyui pack, which wrapped every top-level cv2 function including the CV_* type macros. If you've ever seen CV_16SC3 in old C++ OpenCV code and wondered what the number actually was, this is the node that answers it.

The formula is the same for every depth:

type = depth + (channels - 1) × 8

CV_16S is depth 3, so CV_16SC(1) = 3, CV_16SC(2) = 11, CV_16SC(3) = 19, CV_16SC(4) = 27.

What signed 16-bit is for

int16 arrays are where OpenCV spends a lot of time between operations: gradients, intermediate filter results, and difference images often land in CV_16S because a Sobel or Laplacian of an uint8 image can exceed the 0–255 range. Many functions produce or expect CV_16S intermediates, and a dst/type parameter in this pack may want exactly this constant. That's the realistic use: you're wiring up a filter chain, some step demands the signed 16-bit type code, and instead of hardcoding 19 for 3 channels you compute it with the node.

It's a genuinely niche tool - you're doing classical filter/intermediate math in an OpenCV-native pipeline, not running diffusion. The KB's post-processing essay frames this whole layer as "reach for the cheap deterministic primitive," and this node is the plumbing between those primitives: a constant generator that keeps your type codes honest.

The spec

• channels - INT. 1–4 is the useful range (CV_16SC1=3, CV_16SC2=11, CV_16SC3=19, CV_16SC4=27).
• Output: int - the type constant.

Install

ComfyUI Manager → search opencv-comfyui → install, or:

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

Restart. Requirements: opencv-contrib-python, numpy, torch. No model files.

Anything to watch?

Watch the sibling confusion more than the node itself. CV_16UC_0 (unsigned) has depth 2 and CV_16FC_0 (float) has depth 7; they're different constants for different semantics. Signedness matters to OpenCV - a value meant to hold negative gradients must be CV_16S, not CV_16U. Feed the wrong constant and downstream math silently misbehaves. This node, on its own, is a trivial calculator; getting the kind of constant right is where the actual thinking lives.