Compare Images

Easily compare two images and spot differences

100% Free No installation required Secure & private

Trusted by

Rate this tool

0/5 · 0 votes

Every metadata app in one place

Ten apps for reading, editing, removing, extracting and comparing what is hidden inside your files.

See all metadata apps Open Workspace

Image comparison at a glance

What the app measures, before you add anything.

9 comparison methods to pick from, Butteraugli by default

2 images per comparison, the reference and the one you check

7 image formats accepted, from JPG to TIFF

1 report you can export as a PDF, at no cost

Nine ways to compare images, not one

Compare images with the method that fits the question you are asking. The app offers 9 comparison algorithms, from perceptual measures that model how the eye reads a change to pixel measures that count every altered dot.

Butteraugli is the default. It estimates the difference the way a person would see it, so a change that catches the eye scores high and a change nobody would notice scores low.

The direction matters as much as the number. SSIM, NCC and PSNR say how alike the two images are, so a higher figure is a closer match. Butteraugli, LPIPS, AE, MAE and RMSE say how far apart they are, so a lower figure is the closer match. PDC simply counts the pixels that changed.

The method, the highlight colour and the threshold are set in the second step, before the comparison runs.

Butteraugli is the default. The threshold only applies to the pixel-based methods.

Find the difference between two images

Find the difference between two images and the app hands back a difference map: the second image with every changed area painted in the colour you chose. An object that was removed, a face that was edited, or a logo that moved shows up as a bright patch on an otherwise quiet frame.

There are 8 highlight colours: red, green, blue, yellow, orange, violet, black and white. Red is the default because it reads well on most photographs, but a red-heavy picture is easier to check in green.

For the pixel based methods, the threshold decides how large a change has to be before it counts, on a scale of 0 to 100. It starts at 4, which keeps faint compression noise out of the map while real edits still register.

LPIPS is the one method that returns a score without a difference map. The other 8 return both.

Every changed area is shown in the color you pick, red unless you change it.

Read the image comparison report

Every comparison ends in one report. It opens with a similarity reading for the method you picked, then the difference map, then the numbers the reading was built from.

Methods that work per colour channel report red, green and blue on their own lines as well as a combined figure, so a colour cast shows up as the one channel that moved. The report also states what share of the total pixels differ.

File Details lines up the name, the size, the dimensions and the file type of both images side by side. That is usually where a resize or a second round of compression gives itself away before you have read a single score.

Export PDF saves the whole report as one file, free, with nothing to upgrade to first.

Which method should you compare images with

All 9 are free and all 9 run on the same two images. They differ in what the number means.

Method	What it measures	Higher means more alike
Butteraugli	Perceived visual difference	No
LPIPS	Learned perceptual similarity	No
SSIM	Structural similarity	Yes
AE	Absolute pixel error	No
MAE	Average pixel error	No
NCC	Correlation of pixel values	Yes
PSNR	Signal to noise ratio	Yes
RMSE	Root mean squared error	No
PDC	Count of differing pixels	No

Butteraugli, LPIPS and SSIM return a single fused score, so the colour threshold does not apply to them and the control stays hidden until you switch to one of the pixel based methods. LPIPS is also the only method that returns no difference map. Every other method returns a map and a set of per channel values alongside its score.

When an image comparison earns its keep

Three jobs where two versions of one picture arrive with no note about what changed.

Forensics and authenticity

Two copies of one photograph should match. When they do not, the difference map shows which part of the frame was touched and the score says how far it moved.

Check the Impact of Compression on Image Quality

A smaller file always costs something. Compare the compressed copy against the original and the numbers say what the saving cost, instead of leaving it to the eye.

Design and review rounds

Two exports of one layout rarely differ where you expect. The map points at the pixels that moved, so a review takes a glance rather than a second pair of eyes.

Your images stay yours

The pictures people compare here are often evidence, unreleased work, or somebody else's face. Here is how they are handled.

Encrypted in transit

Both images travel over an encrypted connection. The link that brings the result back is generated at random, so it cannot be guessed by anyone who was not given it.

No human access

Nobody opens, reads or copies the images you compare, because the comparison is fully automated. They are never used to train AI models, and the copyright stays yours.

GDPR compliant

The servers that do the work are located in Europe, run by a named German controller, under a data processing agreement you can download.

How to compare images

Upload

Add the first image. This is the reference the second one is measured against.
Add the second image

Add the image you want to check for changes.
Choose a method

Choose the comparison model.
Compare

Start the comparison. Both images are checked and a similarity score comes back.
Review

View the highlighted differences between the two images.

Compare images FAQ

How can I compare two images?

Add the first image as the reference and the second as the one you want to check, pick a comparison method, and start the comparison. What comes back is a similarity score, a difference map with the changed areas marked in colour, and the numbers behind both. There is nothing to install and no account to open.

How do I find the difference between two images?

Add both images and read the difference map. It shows the second image with every changed area painted in the colour you chose, so a removed object, an edited detail, or a shifted logo is visible at a glance. The report also states what share of the pixels differ, which turns a hunch into a figure.

Can I compare two images for similarity?

Yes. Every comparison returns a score. SSIM, NCC and PSNR are similarity measures, so a higher number means the two images are more alike. Butteraugli, LPIPS, AE, MAE and RMSE are difference measures, so a lower number is the closer match. PDC reports a plain count of the pixels that changed.

Which method should I use to compare images?

Butteraugli is the default and it suits most pictures, because it estimates the difference the way a person would see it. Use SSIM when you want a structural similarity figure between 0 and 1. Use PDC when you want a count of the pixels that changed rather than a judgement about how they look.

Can I compare images that are different sizes?

The comparison lines the two images up pixel for pixel, so it works best when both share the same dimensions. There is no automatic alignment step. The File Details table reports the dimensions of both files, so a mismatch is easy to spot, and scaling the copy back to the original size before you compare gives the cleanest result.

Which image formats can I compare?

JPG, JPEG, PNG, WebP, GIF, BMP and TIFF. The two images do not have to share a format, so a PNG can be compared against a JPG of the same picture.

Can ChatGPT compare images?

An assistant can describe what it sees in two pictures, but it does not measure them, and asking twice can give you two answers. This app runs one of 9 named algorithms, returns a score you can repeat, marks every changed area on a map, and exports the result as a PDF you can hand to somebody else.