OpenCV CV_64FC_0

OpenCV CV_64FC_0 is the double-precision member of this pack's family of type-code helper nodes. Give it a channels count, get back the packed cv2 type constant for a 64-bit float image. CV_64FC(1) returns 6, CV_64FC(3) returns 22. That's the whole job, and it's a job you mostly won't need.

It comes from geroldmeisinger/opencv-comfyui, a pack that auto-generates a ComfyUI node for every top-level standalone cv2 function - about 635 of them. The README is refreshingly honest: the nodes are "ugly and complex to use" and you should expect dragons. CV_64FC_0 is one of the plainest, because all it does is wrap cv2.CV_64FC(channels).

What the number means

OpenCV encodes an image's data type into a single integer: depth + (channels - 1) * 8, where CV_64F (double-precision float) is depth 6. So CV_64FC(1) = 6, CV_64FC(2) = 14, CV_64FC(3) = 22. The _C in the name stands for "channels" - the depth is baked in.

Where this gets real for you: many OpenCV operations that produce fractional results - transforms, distance fields, calibration matrices, and the dct/decompose* nodes elsewhere in this pack - hand back float64 arrays. ComfyUI's own images are float32 tensors scaled 0–1, so the two worlds disagree by design. That's why the pack's conversion nodes (Image2Nparray → Nparrays2Image) exist and why you'll occasionally see a float64 nparray that refuses to behave like a picture. If a function needs a target type constant, CV_64FC_0 is how you build it.

Inputs and outputs

• channels (INT) - the channel count: 1 for scalar/grayscale data, 3 for color.
• int (INT, output) - the packed type code, ready to feed a downstream node's type parameter.

Install and first run

ComfyUI Manager can install it by searching opencv-comfyui; otherwise:

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

Restart ComfyUI afterwards. No models, no weights - just numpy, torch, and the OpenCV contrib build that the requirements.txt pins (note: the README's pip install opencv-python-contrib is the wrong name; the real package is opencv-contrib-python).

Where people trip

The most common failure with this node family isn't the node itself - it's treating every nparray like a viewable image. The pack README's Nparrays2Image error - 'NoneType' object has no attribute 'shape' - happens exactly when you feed it a float vector or matrix that was never meant to be pixels. A float64 type code is often the smell: if the values aren't in 0–255 uint8 territory, it's not a picture, it's data. Check what the OpenCV function actually returns before wiring it to an image preview.

And the _0/_1 suffix thing again: cv2's Python stubs declare most functions twice (once for MatLike, once for UMat), and the code generator dutifully makes a node per overload. They behave identically here, so don't agonize over which variant you drop in.