CV Quality Compare (cv2.quality)

Scores how close ONE image is to a reference (cv2.quality) - the objective way to compare two settings of a filter, an upscaler or a denoiser instead of eyeballing them. Outputs the overall score and a per-pixel quality map showing WHERE the images disagree. Read the direction carefully: for SSIM and PSNR higher is better, for MSE and GMSD LOWER is better ('higher_is_better' says which). Both images must be the same size. To score a whole BATCH of candidates against one reference, use 'CV Quality Compare Batch' instead - it computes the reference's state once and returns the winner's index.

By bmad4ever·Created 3 months ago·Updated 2 days ago· 0

reference
image

score
higher_is_better
quality_map

◄metricSSIM (structural similarity)►

Categoryimage/CV/quality

Inputs (3)

Name	Type	Default	Description
reference	NPARRAY,IMAGE,MASK		The ground-truth / original image. Accepts a ComfyUI IMAGE/MASK directly (frame 0 of a batch) or an NPARRAY. Arithmetic ops (add, multiply, etc.) process the full IMAGE batch when both inputs have the same batch size.
image	NPARRAY,IMAGE,MASK		The processed image to score against the reference. Must be the same size. Accepts a ComfyUI IMAGE/MASK directly (frame 0 of a batch) or an NPARRAY. Arithmetic ops (add, multiply, etc.) process the full IMAGE batch when both inputs have the same batch size.
metric	COMBO	SSIM (structural similarity)	SSIM tracks perceived structural change (0-1, 1 = identical) and is the usual choice. PSNR is a log-scale error in dB (>40 is typically indistinguishable). MSE is the raw squared error. GMSD compares gradient structure and correlates well with perceived sharpness loss.

Outputs (3)

Name	Type	Description
score	FLOAT	Overall score, averaged over the colour channels.
higher_is_better	BOOLEAN	True for SSIM/PSNR, False for MSE/GMSD. Wire it into a comparison node so a workflow can pick the better of two candidates without hardcoding the metric's direction.
quality_map	NPARRAY	Per-pixel quality/error map - preview it with 'Preview CV Array' (normalize/heatmap) to see WHERE the two images differ. GMSD's map is HALF the input's resolution.