Triage before a download
The background remover fetches between 45 and 192 MB. Check the photo first and find out whether it is worth spending.
Private image tools
Find out whether a photo will cut out, before you download anything
Three measurements, taken on your own device in a fraction of a second: how sharp the photo is, how much of it has been clipped to pure black or pure white, and how hard its edges will be to cut along. Each one comes back with a sentence saying what it means, not just a number, and the result prints its own measured run time so you can see what it cost.
This one is not AI. It is Laplacian variance, a clipped-pixel count and two edge measurements: arithmetic over the pixels you supplied. Nothing was trained, nothing is fetched, and it cannot tell you what is in the picture. It can tell you whether the background remover is going to have a hard time with it.
Result proof
Three scores you can check against the raw numbers
01
Input
Any image file your browser can decode
02
Local pass
Laplacian variance, clipped-histogram fractions, and Sobel edge measurements at a fixed 768 pixel working size
03
Result
Three scores and a plain-language verdict
How to verify it: Every score is shown beside the measurement it came from, so you can disagree with the mapping and still use the fact. Run it on a photo you know is sharp and one you know is blurred and compare the variance; then run the background remover on both and see whether the outlook matched.
What people use this for
Picking the best of a burst
Run it across several frames of the same shot and keep the one with the highest sharpness rather than judging by eye on a small screen.
Explaining a bad cutout
When a cutout comes out fuzzy, the measurements usually say why: a soft edge, a clipped shadow, or a background that shares the subject colours.
Checking a supplier photo
Before you build a listing around an image someone sent you, find out whether it can be cut out at all.
Limits worth knowing
- It measures the whole frame, not the subject. A sharp subject against a deliberately blurred background scores lower than the subject deserves.
- The border-versus-centre colour comparison assumes the subject is nearer the middle than the edges. A subject that fills the frame or sits in a corner will be reported as harder than it is.
- The band boundaries are a convention taken from standard blur-detection practice, not a threshold learned from a labelled set of photographs. The raw numbers are shown so you can draw the line elsewhere.
- It cannot tell a photo that is meant to be soft from one that failed. Grain, film look and intentional motion all read as low sharpness.
- It predicts difficulty. It does not remove a background, sharpen anything, or change your file in any way.
How the local workflow runs
- Choose a photo. It stays on your device and is measured as soon as it decodes.
- Read the cutout outlook first: that is the line that decides whether to go on.
- Read the three cards. Each one names what was measured and what it means for a cutout.
- If sharpness or noise is the weak one, run the deblur or denoise tool and check again.
- When the outlook is good enough, go to the background remover.
Frequently asked questions
Is this AI?
No. Three classical measurements over the pixels you supplied: the variance of the Laplacian for sharpness, the fraction of pixels sitting at pure black or pure white for exposure, and two edge measurements for cutout difficulty. Nothing was trained, nothing is downloaded, and the tool has no idea what is in your photo. It does not know whether it is looking at a person, a product or a receipt.
Why would I run this before removing a background?
Because the background remover is a 45 to 192 MB download, and the two properties that decide whether it produces a clean cutout are measurable in a fraction of a second for free. If the edges in your photo are soft ramps rather than clean turns, no matting model will find a crisp boundary in them, and knowing that before you spend the download is the entire point.
What does the sharpness number actually mean?
It is the variance of the Laplacian: a filter that responds to how quickly brightness changes from one pixel to the next. A photo with real fine detail produces a wide spread of those responses and a high variance; a blurred one produces a narrow spread and a low variance. The raw variance is shown beside the score, because the score is a mapping and the variance is the fact.
Why is the score the same on my phone and my laptop?
Because the measurements are taken at a fixed working size of 768 pixels on the longest edge rather than at your device limit. Sharpness measured in pixels changes with resolution, so the same photograph would score differently on a phone and a desktop if the analysis followed the device. A score that moves with the hardware is not a score.
What is "cutout difficulty" measuring?
Two things, combined. First, how abruptly the strongest edges in the frame turn over: a clean edge concentrates its gradient into about a pixel, while motion blur, shallow depth of field and fine hair spread it across a ramp several pixels wide. Second, how far the colours near the frame border sit from the colours in the middle. A subject that shares its palette with its background is hard to separate by any means.
When is the tool wrong?
The border-versus-centre comparison assumes your subject is nearer the middle of the frame than the edges. That holds for the product and portrait shots this site is mostly used for, and it is false for a subject that fills the frame or sits in a corner, where the tool will report the cutout as harder than it is. Sharpness is also a whole-frame measurement: a sharp subject against a deliberately blurred background scores lower than the subject deserves.
Does my photo leave my device?
No. The file is decoded by your browser and measured by your browser. There is no upload step, no server, and no model fetch, so there is nothing for this tool to send anywhere even if it wanted to.
Did this exist before?
A version of it did, called Quality Score, and it failed in an instructive way: it computed a correct answer and dropped it, because nothing on the site could display a text result. The measurements are different now and aimed at cutouts specifically, and the result has two real places to appear: this page, and the image editor panel.
Related tools
Background Remover
The tool this one exists to triage for.
Photo Denoise
Clean grain and JPEG blocking out before cutting out. Also no download.
AI Deblur
Recover a mild focus miss or camera shake, optionally with a real model.
Where it struggles
The cases background removal gets wrong, why, and what to do instead.
Ready to try it?
No upload, no model download, and nothing is changed about your file.
Check a photo →