OpenCV getOptimalNewCameraMatrix_1

OpenCV getOptimalNewCameraMatrix_1 is a duplicate of getOptimalNewCameraMatrix_0. The pack's generator walks OpenCV's type stubs, numbers each overload, and cv2.getOptimalNewCameraMatrix has two overloads that flatten to the same signature - so you get two byte-identical nodes. Same six inputs, same two outputs, same behavior. If this is the one you found, everything that follows applies; there is no hidden difference in the _1 version.

What it does (and why you care)

This is the pre-undistortion calculator: feed it your camera matrix, distortion coefficients, and source image size, and it returns an adjusted camera matrix plus the ROI of valid pixels, so undistort can straighten your lens-distorted photo without leaving a border of dead black pixels. The alpha input is the whole knob:

• alpha = 0 - crop to valid pixels only: no borders, maximum resolution, edges get trimmed.
• alpha = 1 - keep every original pixel: black borders appear.
• In between - the trade-off. 0.5 is a reasonable starting point.

The inputs

• cameraMatrix (NPARRAY) and distCoeffs (NPARRAY) - calibration data as OpenCV numpy arrays.
• imageSize (STRING) - source size as a literal, [1920, 1080] style. Brackets matter.
• alpha (FLOAT) - free-scaling parameter, 0–1.
• newImgSize (STRING) - optional output size literal, [] to match source.
• centerPrincipalPoint (BOOLEAN) - usually False here.

Outputs are nparray (the tuned 3×3 camera matrix, for undistort) and literal (STRING) - the valid-pixels ROI as a text tuple like (x, y, w, h). That second output is the pack's classic composite-type wart: the ROI is a Rect, so it comes out serialized as a string you'll have to parse back into numbers for a crop node. Annoying, but optional - undistortion works with the matrix alone; the ROI just tells you the honest crop.

Install

Standard for the pack - ComfyUI Manager, search opencv-comfyui (display "OpenCV"), or:

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

Restart, no models.

Pack-wide gotchas apply: image data travels as NPARRAY (numpy, BGR) with batch size 1 via Image2Nparray/Nparrays2Image, and every Size literal needs brackets or you'll hit invalid syntax (<unknown>, line 0). And the standing advice for duplicates: pick one variant, standardize, and don't diff them looking for differences that don't exist.