เปรียบเทียบรูปภาพสองภาพและดูความแตกต่างได้ง่ายๆ
หรือวาง ลากไฟล์มาวาง หรือเลือกจากคลาวด์
หรือวาง ลากไฟล์มาวาง หรือเลือกจากคลาวด์
What the app measures, before you add anything.
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.
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 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.
All 9 are free and all 9 run on the same two images. They differ in what the number means.
| วิธี | What it measures | Higher means more alike |
|---|---|---|
| Butteraugli | Perceived visual difference | ไม่ใช่ |
| LPIPS | Learned perceptual similarity | ไม่ใช่ |
| SSIM | Structural similarity | ใช่ |
| AE | Absolute pixel error | ไม่ใช่ |
| MAE | Average pixel error | ไม่ใช่ |
| NCC | Correlation of pixel values | ใช่ |
| PSNR | Signal to noise ratio | ใช่ |
| RMSE | Root mean squared error | ไม่ใช่ |
| PDC | Count of differing pixels | ไม่ใช่ |
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.
Three jobs where two versions of one picture arrive with no note about what changed.
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.
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.
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.
The pictures people compare here are often evidence, unreleased work, or somebody else's face. Here is how they are handled.
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.
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.
The servers that do the work are located exclusively in Europe, run by a named German controller, under a data processing agreement you can download.
Add the first image. This is the reference the second one is measured against.
Add the image you want to check for changes.
เลือกโมเดลการเปรียบเทียบ
Start the comparison. Both images are checked and a similarity score comes back.
ดูส่วนที่ถูกไฮไลต์ซึ่งแสดงความแตกต่างระหว่างรูปทั้งสอง
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.
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.
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.
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.
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.
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.
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.
Yes. Comparing images is free and no account is needed. Each image can be up to 75 MB, and a paid plan raises that to 8 GB.
Add both images here and they open in the workspace, where the method is chosen and the report is built.
The reference image first, then the image you want to check. One step each, from your device, a link, or cloud storage.
Pick one of 9 algorithms, one of 8 highlight colours, and how large a change has to be before it counts.
One picture with the changed areas painted in, next to the score, the per channel values, and the share of pixels that differ.
The score, the numbers and the file details of both images, saved as one file at no cost.
A walkthrough of the 2 upload steps and how to read the score that comes back.
อ่านบทความ
The same idea for moving pictures: frame by frame analysis and a quality score for 2 videos.
อ่านบทความ
Compare videos and PDF files the same way, and look inside an image while you are here.
