BlazeDiff vs ssim.js

@blazediff/ssim is ~21% faster than ssim.js on standard SSIM, and the native @blazediff/ssim-native is ~19.8x faster. Both take RGBA pixels and return a similarity score from 0 to 1. ssim.js is archived on GitHub, and its last release was 3.5.0 in 2020.

ssim.js is a JavaScript port of the Wang et al. MATLAB reference. It offers several variants, and its default (weber) trades accuracy for speed. BlazeDiff ships standard SSIM that matches MATLAB ssim.m with an error below 0.01%, plus MS-SSIM and Hitchhiker’s SSIM.

At a glance

	ssim.js	BlazeDiff
License	MIT	MIT
Status	Archived, last release 2020	Maintained
Runtimes	Node, browsers	Node, browsers, Deno, Bun (`ssim`); native Node (`ssim-native`)
Input	`ImageData`-like `{ data, width, height }`	RGBA buffers (`ssim`); file paths or encoded bytes (`ssim-native`)
Variants	`original`, `fast`, `bezkrovny`, `weber`	SSIM, MS-SSIM, Hitchhiker’s SSIM; perceptual SSIM in native
Score map	Yes (`ssim_map`)	Yes (grayscale map image)
Pass or fail result	No, you compare the score	`ssim-native` takes `minScore` and returns `match`

Benchmarks

The first chart compares standard SSIM against ssim.js fast. The second compares Hitchhiker’s SSIM against ssim.js weber, its fastest mode. Decode is excluded.

Fast Original ( `@blazediff/ssim` using `ssim` vs `ssim.js` using `fast` algorithm) (image IO excluded)

TOTAL TIME · 22 FIXTURES · MEAN OF 25 RUNS · IO EXCLUDED

BlazeDiffOther librariesLower is better

Apple M1 Max · darwin/arm64 · Node v22.14.0 · rustc 1.97.1 (8bab26f4f 2026-07-14) · 2026-10-22 · 234b300-dirty · pnpm benchmark ssim

Per-fixture results (22)

Fixture	ssim.js	BlazeDiff	BlazeDiff (ssim-native)	BlazeDiff vs ssim.js	BlazeDiff (ssim-native) vs ssim.js
blazediff/1	92.94ms	65.98ms	5.36ms	1.41x	17.33x
blazediff/1 (identical)	99.29ms	65.49ms	7.92ms	1.52x	12.54x
blazediff/2	39.45ms	23.49ms	1.55ms	1.68x	25.42x
blazediff/2 (identical)	36.44ms	24.04ms	1.60ms	1.52x	22.77x
blazediff/3	112ms	98.71ms	5.47ms	1.13x	20.44x
blazediff/3 (identical)	106ms	100ms	5.03ms	1.06x	21.14x
blazediff/4	259ms	226ms	12.26ms	1.15x	21.10x
blazediff/4 (identical)	268ms	229ms	11.99ms	1.17x	22.31x
pixelmatch/1	28.25ms	20.13ms	1.63ms	1.40x	17.29x
pixelmatch/1 (identical)	30.38ms	19.79ms	1.57ms	1.54x	19.34x
pixelmatch/2	13.92ms	10.90ms	0.87ms	1.28x	16.05x
pixelmatch/2 (identical)	14.40ms	9.96ms	0.76ms	1.45x	19.06x
pixelmatch/3	30.00ms	19.92ms	1.60ms	1.51x	18.74x
pixelmatch/3 (identical)	28.97ms	19.86ms	1.63ms	1.46x	17.73x
pixelmatch/4	19.65ms	11.76ms	0.79ms	1.67x	24.77x
pixelmatch/4 (identical)	20.60ms	11.88ms	0.81ms	1.73x	25.50x
pixelmatch/5	14.34ms	10.15ms	0.76ms	1.41x	18.93x
pixelmatch/5 (identical)	15.07ms	9.94ms	0.78ms	1.52x	19.33x
pixelmatch/6	14.38ms	9.98ms	0.76ms	1.44x	18.95x
pixelmatch/6 (identical)	13.79ms	10.70ms	0.77ms	1.29x	17.89x
pixelmatch/7	27.20ms	16.64ms	1.32ms	1.63x	20.65x
pixelmatch/7 (identical)	27.96ms	16.63ms	1.10ms	1.68x	25.41x
TOTAL	1.31s	1.03s	66.34ms	1.27x	19.77x

Hitchhikers SSIM SSIM (`@blazediff/ssim` using `hitchhikers-ssim` vs `ssim.js` using `weber` algorithm) (image IO excluded)

TOTAL TIME · 22 FIXTURES · MEAN OF 25 RUNS · IO EXCLUDED

BlazeDiffOther librariesLower is better

Apple M1 Max · darwin/arm64 · Node v22.14.0 · rustc 1.97.1 (8bab26f4f 2026-07-14) · 2026-10-22 · 234b300-dirty · pnpm benchmark hitchhikers-ssim

Per-fixture results (22)

Fixture	ssim.js	BlazeDiff	BlazeDiff (ssim-native)	BlazeDiff vs ssim.js	BlazeDiff (ssim-native) vs ssim.js
blazediff/1	77.27ms	12.38ms	2.92ms	6.24x	26.42x
blazediff/1 (identical)	77.43ms	12.44ms	3.06ms	6.22x	25.33x
blazediff/2	35.29ms	10.04ms	2.33ms	3.52x	15.16x
blazediff/2 (identical)	40.30ms	10.45ms	2.50ms	3.86x	16.12x
blazediff/3	123ms	47.14ms	10.91ms	2.62x	11.31x
blazediff/3 (identical)	125ms	47.55ms	11.03ms	2.62x	11.29x
blazediff/4	302ms	112ms	25.70ms	2.69x	11.73x
blazediff/4 (identical)	316ms	110ms	25.09ms	2.88x	12.59x
pixelmatch/1	22.72ms	3.75ms	0.84ms	6.07x	27.10x
pixelmatch/1 (identical)	23.13ms	3.91ms	0.88ms	5.91x	26.17x
pixelmatch/2	11.31ms	1.88ms	0.40ms	6.02x	28.14x
pixelmatch/2 (identical)	11.50ms	1.91ms	0.40ms	6.01x	28.48x
pixelmatch/3	22.96ms	3.75ms	0.84ms	6.12x	27.40x
pixelmatch/3 (identical)	23.90ms	3.75ms	0.84ms	6.38x	28.36x
pixelmatch/4	18.61ms	5.15ms	1.16ms	3.61x	15.97x
pixelmatch/4 (identical)	18.70ms	5.44ms	1.18ms	3.44x	15.88x
pixelmatch/5	11.38ms	1.92ms	0.41ms	5.94x	27.75x
pixelmatch/5 (identical)	11.78ms	1.86ms	0.40ms	6.32x	29.19x
pixelmatch/6	11.37ms	1.96ms	0.43ms	5.81x	26.69x
pixelmatch/6 (identical)	11.91ms	1.89ms	0.40ms	6.29x	29.62x
pixelmatch/7	24.87ms	8.65ms	1.71ms	2.87x	14.55x
pixelmatch/7 (identical)	25.05ms	7.20ms	1.64ms	3.48x	15.24x
TOTAL	1.34s	414ms	95.09ms	3.24x	14.14x

Hitchhiker’s SSIM is ~3.2x faster than ssim.js weber in JavaScript.

Switching

ssim.js takes two image objects. @blazediff/ssim takes the two pixel buffers and the size, in the same order as a pixel diff.


// Before
import ssim from "ssim.js";
const { mssim } = ssim(img1, img2);
 
// After
import ssim from "@blazediff/ssim/ssim";
const score = ssim(img1.data, img2.data, undefined, img1.width, img1.height);

Scores from different SSIM variants are not interchangeable. If you gated tests on an ssim.js weber score, measure your fixtures again before you copy the threshold.

When to keep ssim.js

You depend on the bezkrovny or weber variant scores exactly.
You use its UQI mode (k1 and k2 set to 0).