OpenCV compareHist_0

Sometimes you don't care about pixels at all - you care about vibe. Is this generated image closer in color to reference A or reference B? Does this frame match the style of the ones before it? That's histogram territory, and compareHist_0 is the OpenCV function cv2.compareHist with a ComfyUI face on it. Feed it two histograms, get back a single float that tells you how much they agree.

What it is

One of the auto-generated opencv-comfyui nodes, and it's the purest kind: no image in, no image out. Inputs are two histograms (H1, H2), the output is a FLOAT score. It's a measurement node, the kind you wire into a comparison or a conditional rather than straight into a preview.

How it works

A histogram is just a distribution - "how much of this image sits in each brightness/color bin." compareHist measures how far apart two distributions are, and which method changes what the score means:

• 0 - CORREL: higher = more similar (-1..1)
• 1 - CHISQR: lower = more similar (0 = identical)
• 2 - INTERSECT: higher = more similar (raw bin overlap)
• 3 - BHATTACHARYYA: lower = more similar (0 = identical)

If you're doing "is this like that", 2 (intersect) is the intuitive one - it's literally how much the histograms overlap. 0 and 3 are the ones the image-retrieval crowd actually uses, because they're normalized and behave better across brightness changes.

The honest caveat: this pack can't make histograms

Here's where people get burned. compareHist_0 compares histograms, but the pack has no calcHist node - I grepped the source, it isn't there (its return type wasn't supported by the generator). So you can't build the H1/H2 inputs inside opencv-comfyui itself. You need them from elsewhere: a tiny custom Python node running cv2.calcHist, or any node that outputs an array you can treat as a distribution.

That makes this one of the most "you probably won't use it" nodes in the pack - unless you already have a histogram source, in which case it's a genuinely handy similarity score for duplicate detection or checking that a batch of renders didn't drift in color.

The inputs and outputs that matter

• H1 / H2 (NPARRAY) - the two histograms. Same size and type, or expect an assertion.
• method (INT) - pick from the list above; 2 to start.
• Output: float - the score. This is not an image. Do not wire it into Nparrays2Image; you'll get the README's 'NoneType' object has no attribute 'shape' because it's a scalar, not an array. Feed it into a math or logic node instead.

Install and gotchas

Same pack install as everything else - ComfyUI Manager, search "opencv-comfyui", or git clone https://github.com/geroldmeisinger/opencv-comfyui into ComfyUI/custom_nodes. Requires opencv-python-contrib. Watch for the guidedFilter import conflict if you have multiple OpenCV installs. And remember the golden rule of this pack: any node whose output you try to render as an image will remind you the hard way that some functions just return numbers.