Photo Check

Photo Check measures three things about a photograph and tells you, in a sentence, what each one means for the cutout you are about to attempt. It runs on your device in a fraction of a second, and it prints its own measured run time beside the result. Nothing is uploaded and nothing is downloaded, because there is no model.

It exists because the background remover is a 45 to 192 MB download and the properties that decide whether it will work are free to measure.

This is not AI, and it matters which one it is

Three classical measurements over the pixels you supplied:

• Sharpness is the variance of the Laplacian. The Laplacian is 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 blurred one produces a narrow spread.
• Exposure is the fraction of pixels sitting at or beyond pure black and pure white. Those pixels have no detail left to recover.
• Cutout difficulty combines two edge measurements: how abruptly the strongest edges in the frame turn over, and how far the colours near the frame border sit from the colours in the middle.

Nothing was trained. The tool does not know whether it is looking at a person, a product or a receipt, and it will tell you nothing about the content of your picture.

Why the working size is fixed

Every measurement is taken at 768 pixels on the longest edge, whatever your file is. That is deliberate and it is the difference between a score and a number.

Sharpness measured in pixels is resolution dependent: a blur of a fixed physical width covers more pixels in a larger image, so the same photograph measures differently at 4096 pixels than at 1536. The rest of this site caps images by device, which would mean a phone and a laptop scoring the same file differently. A score that moves with the hardware is not a score.

The fixed size also bounds what this tool costs, and the number is the same whether you hand it a phone snap or a 48 megapixel raw conversion. At the largest working size it uses, three buffers of about 2.4 MB each are live at once, so the measurement runs in roughly 7 MB. The background remover, by comparison, works at your device's full resolution.

Reading the three cards

Each card carries a band, a score out of 100 and a sentence. The band is a convention, the score is a mapping, and the raw measurement underneath is the fact. All three are shown, in that order of reliability.

Sharpness

A high variance means there is real detail for a cutout edge to follow. A low one means there is not, and no background remover can find an edge the camera never recorded. If this is your weak card, Photo Denoise will not help, because noise is not the problem; AI Deblur might, if the softness is mild camera shake or a small focus miss.

Exposure

The tool distinguishes blown highlights from crushed shadows and names whichever is worse. Both matter for the same reason: wherever the subject meets a fully clipped area, there is no boundary left in the file to find. A product against a blown white backdrop is the common case, and the fix is a re-shoot or a raw file, not a filter.

Around one percent clipped at one end is normal on a real photograph with a specular highlight, so the score falls gently at first and steeply once a tenth of the frame has lost its detail.

Cutout difficulty

This is the card that predicts the result. It says one of three things:

• The edges are wide ramps rather than clean turns. This is what motion blur, shallow depth of field and fine hair all look like to a measurement. Expect a soft band around the subject.
• The border colours are close to the centre colours. The subject and its background are hard to tell apart by colour alone, so expect the model to lose parts of the subject.
• Neither. Edges turn over cleanly and the border differs from the middle, which are the two conditions a background remover needs.

When the numbers are right and the conclusion is wrong

Three cases, and they are the reason the tool shows you what it measured rather than only its verdict.

A subject that fills the frame, or sits in a corner. The border-versus-centre comparison assumes your subject is nearer the middle than the edges. That is true of the product and portrait shots this site is mostly used for and false of a tight crop, where the tool will report the cutout as harder than it is.

A sharp subject against a deliberately blurred background. Sharpness is a whole-frame measurement, so a portrait shot at a wide aperture scores lower than the subject deserves. The cutout card is the one to read in that case, and it will often disagree with the sharpness card. That disagreement is information.

A photo that is meant to be soft. Grain, film look and intentional motion all read as low sharpness. The tool cannot tell a deliberate choice from a failure and does not try to.

What it does not do

It does not change your file. It does not remove a background, sharpen anything, or write anything to disk. It reads an image and reports numbers.

It also will not tell you the tool is unusable. Every band boundary here is a convention taken from standard blur-detection practice rather than a threshold learned from a labelled set of photographs, so a "poor" is a prompt to look rather than a verdict. The raw measurement is printed beside every score precisely so you can draw the line somewhere else.

Where to go next

• Background Remover when the outlook is good.
• Photo Denoise when grain is the problem. Also free and also local.
• AI Deblur for a mild focus miss.
• Where background removal struggles for the cases that are hard for reasons no measurement catches.