A lightweight image comparison tool

Build Status Coveralls Coverage Code Climate Grade

NPM version NPM License


Coverage Report API Documentation

Gitter Support

Table of Contents

Image Comparison and Result



Install this module with the following command:

npm install blink-diff

Add the module to your package.json dependencies:

npm install --save blink-diff

Add the module to your package.json dev-dependencies:

npm install --save-dev blink-diff


The package can be used in two different ways:

Command-Line usage

The command-line tool can be found in the bin directory. You can run the application with

blink-diff --output <output>.png <image1>.png <image2>.png

Use image1 and image2 as the images you want to compare. Only PNGs are supported at this point.

The command-line tool exposes a couple of flags and parameters for the comparison:

--verbose           Turn on verbose mode
--debug             Turn on debug mode - leaving all filters and modifications on the result
--threshold p       Number of pixels/percent 'p' below which differences are ignored
--threshold-type t  'pixel' and 'percent' as type of threshold. (default: pixel)
--delta p           Max. distance colors in the 4 dimensional color-space without triggering a difference. (default: 20)
--copyImageA        Copies first image to output as base. (default: true)
--copyImageB        Copies second image to output as base.
--no-copy           Doesn't copy anything to output as base.
--output o          Write difference to the file 'o'
--filter f          Filters f (separated with comma) that will be applied before the comparison.
--no-composition    Turns the composition feature off
--compose-ltr       Compose output image from left to right
--compose-ttb       Compose output image from top to bottom
--hide-shift        Hides shift highlighting (default: false)
--h-shift           Acceptable horizontal shift of pixel. (default: 0)
--v-shift           Acceptable vertical shift of pixel. (default: 0)
--block-out x,y,w,h Block-out area. Can be repeated multiple times.
--version           Print version
--help              This help

Object usage

The package can also be used directly in code, without going through the command-line.


var diff = new BlinkDiff({
    imageAPath: 'path/to/first/image', // Use file-path
    imageBPath: 'path/to/second/image',

    thresholdType: BlinkDiff.THRESHOLD_PERCENT,
    threshold: 0.01, // 1% threshold

    imageOutputPath: 'path/to/output/image'
}); (error, result) {
   if (error) {
      throw error;
   } else {
      console.log(diff.hasPassed(result.code) ? 'Passed' : 'Failed');
      console.log('Found ' + result.differences + ' differences.');

All the parameters that were available in the command-line tool are also available through the class constructor, however they might use slightly different wording. The class exposes additional parameters that are not available from the command-line:


var firstImage = PNGImage.readImage('path/to/first/image', function (err) {

  if (err) {
    throw err;

  var diff = new BlinkDiff({
      imageA: srcImage, // Use already loaded image for first image
      imageBPath: 'path/to/second/image', // Use file-path to select image

      delta: 50, // Make comparison more tolerant

      outputMaskRed: 0,
      outputMaskBlue: 255, // Use blue for highlighting differences

      hideShift: true, // Hide anti-aliasing differences - will still determine but not showing it

      imageOutputPath: 'path/to/output/image'
  }); (error, result) {
    if (error) {
      throw error;
    } else {
      console.log(diff.hasPassed(result.code) ? 'Passed' : 'Failed');
      console.log('Found ' + result.differences + ' differences.');


Images can be cropped before they are compared by using the cropImageA or cropImageB parameters. Single values can be left off, and the system will calculate the correct dimensions. However, x/y coordinates have priority over width/height as the position are usually more important than the dimensions - image will also be clipped by the system when needed.

Perceptual Comparison

The perceptual comparison mode considers the perception of colors in the human brain. It transforms all the colors into a human perception color-space, which is quite different to the typical physical bound RGB color-space. There, in the perceptual color-space, the distance between colors is according to the human perception and should therefore closer resemble the differences a human would perceive seeing the images.


By default, the logger doesn't log events anywhere, but you can change this behavior by overwriting blinkDiff.log:

var blinkDiff = new BlinkDiff({

blinkDiff.log = function (text) {
    // Do whatever you want to do



Sometimes, it is necessary to block-out some specific areas in an image that should be ignored for comparisons. For example, this can be IDs or even time-labels that change with the time. Adding block-outs to images may decrease false positives and therefore stabilizes these comparisons.

The color of the block-outs can be selected by the API parameters. However, the block-out areas will not be visible by default - they are hidden even though they are used. To make them visible, turn the debug-mode on.


There are some examples in the examples folder, in which I used screenshots of YDN to check for visual regressions (and made some manual modifications to the dom to make differences appear ;-)). You can find examples for:

All screenshots were compared to YDN.png, a previously approved screenshot without a regression. Each of the regressions has the screenshot and the output result, highlighting the differences.


Generate the documentation with following command:

npm run docs

The documentation will be generated in the docs folder of the module root.


Run the tests with the following command:

npm run test

The code-coverage will be written to the coverage folder in the module root.

Project Focus

There are three types of image comparisons:

Blink-Diff was initially created to compare screenshots. These images are generally low-frequency, meaning larger areas with the same color and less gradients than in photos. The pixel-by-pixel comparison was chosen as it will trigger for differences that a human might not be able to see. We believe that a bug is still a bug even if a human won't see it - a regression might have happened that wasn't intended. A perceptual comparison would not trigger small differences, possibly missing problems that could get worse down the road. Pixel-by-pixel comparisons have the reputation of triggering too often, adding manual labor, checking images by hand. Blink-Diff was created to keep this in mind and was optimized to reduce false-positives by taking sub-pixeling and anti-aliasing into account. Additional features like thresholds and the pythagorean distance calculation in the four dimensional color-space makes sure that this won't happen too often. Additionally, filters can be applied to the images, for example to compare luminosity of pixels and not the saturation thereof. Blink-Diff also supports partially the perceptual comparison that can be turned on when supplying perceptual=true. Then, the colors will be compared in accordance with the human perception and not according to the physical world. High-frequency filters, however, are not yet supported.

Project Naming

The name comes from the Blink comparator that was used in Astronomy to recognize differences in multiple photos, taking a picture of the same area in the sky over consecutive days, months, or years. Most notably, it was used to discover Pluto.


Feel free to create an issue or create a pull-request if you have an idea on how to improve blink-diff. We are pretty relaxed on the contribution rules; add tests for your pull-requests when possible, but it is also ok if there are none - we'll add them for you. We are trying to improve blink-diff as much as possible, and this can only be done by contributions from the community.

Also, even if you simply gave us an idea for a feature and did not actually write the code, we will still add you as the Contributor down below since it probably wouldn't be there without you. So, keep them coming!


Third-party libraries

The following third-party libraries are used by this module:




The MIT License

Copyright 2014-2015 Yahoo Inc.