cv2.cubeRoot

cv2.cubeRoot computes the cube root of a number. One FLOAT widget, one FLOAT out, no image anywhere. It is the smallest node in bmad4ever's ComfyUI CV pack, and it exists for a reason worth understanding: the pack generates wrappers for every top-level cv2 function it can parse, and cv2's API includes a handful of scalar maths helpers. Some of those wrappers are genuinely useful (cv2.getNumberOfCPUs, cv2.getOptimalDFTSize). This one is the other case.

So why would you wire it? Because every so often a number in a graph needs a cube root, and doing it here keeps the graph self-documenting instead of burying it in an expression node - and because cbrt is defined for negatives, which pow(x, 1/3) is not. std::cbrt(-8) is -2; the same computation via a fractional power returns NaN. If your pipeline is handling log/signed data, that difference is the whole reason the function exists.

What it actually does

Returns the real cube root of val, the single required FLOAT input. Negative inputs come back negative, zero comes back zero, and there's no complex result to surprise you. Values are not clipped to any range - the widget accepts roughly ±1e38, and the cube root of a big number is still a big number.

The output socket is named float and is a plain ComfyUI FLOAT, so it plugs into any FLOAT input in any pack, once you've converted that widget to an input. That's the mechanic worth knowing generally: right-click a numeric widget → Convert widget to input, then wire. A scalar node like this is only useful when something downstream has a socket to receive it.

The honest assessment

You will probably never use this node, and that's not a criticism of it. It's a byproduct of the pack's design - the README is upfront that the raw wrappers are auto-generated and uncurated, and that a generator parsing the type stubs cannot tell which hundred cv2 functions anyone actually wants. cv2.cubeRoot is in the same category as the pack's cv2.currentUIFramework and cv2.CV_16FC: legitimate cv2 API, near-zero workflow value.

Two places it can earn its keep, both rare:

• Tone/gamma maths in linear space. A cube-root is a gamma of 1/3. If you're converting between linear light and a perceptual-ish encoding by hand and want the exponent applied to a scalar control value rather than an image, this is a one-node way to do it.
• Robust statistics. Cube roots of signed quantities (a skewness measure, a signed sum of residuals) come up when you're post-processing arrays this pack already produces - differences, moments, alignments. The step is scalar, the array version needs cv2.pow on an NPARRAY.

If what you actually wanted was "apply a gamma to an image", this is not the node: cv2.LUT with a built table, the intensity-transform wrappers (cv2.intensity_transform_gammaCorrection), or cv2.pow on the array are the real answers. See the deterministic-pixel-operations layer for the wider picture - this is the scalar corner of it.

Installing it

ComfyUI Manager → search the pack title (ComfyUI CV) → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/bmad4ever/comfyui_cv
pip install "opencv-contrib-python-headless~=5.0.0.93"

Python ≥ 3.12 and a V3-API ComfyUI, since the whole pack is written against the V3 node schema. That single pinned dependency is all it needs - but the wheel is the contrib one, and if you install a non-contrib opencv-python on top later, the shared site-packages/cv2 loses its contrib submodules and a chunk of this pack's nodes disappear. tools/repair_opencv_contrib.py --check then --apply is the repair.

Common issues and troubleshooting

The node "does nothing". It computes a number; there's no UI feedback and no preview. Check the value with a Preview as Text / show-value node, or wire it into something that displays.

You can't type a value in because the field is empty. The default is 0 and the widget is a plain float field - click it. The confusion usually comes from having converted the output into a socket elsewhere and expecting this widget to follow.

You wanted this on an image. Wrong node, wrong socket type. FLOAT in, FLOAT out; the array path is cv2.pow.