Checking a Cutout You Did Not Make, Before You Build On It

Somebody else made the cutout. A freelancer delivered a folder of them, a supplier sent product images with the backgrounds already removed, an asset pack promised transparent PNGs, or a client forwarded files their previous agency produced. You are about to build a catalogue, a deck, or a storefront on top of them.

Checking them takes about a minute each and it is worth doing, because the defects in a bad cutout are invisible on a white page and obvious on a dark one, which is to say they surface after everything is built.

Here is the order to check in, and what a failure at each stage actually means.

1. Does it have transparency at all

Start here, because this failure is more common than it should be and it invalidates everything else.

Drop the file into Check Transparency. It reads every pixel rather than a sample, and it tells you exactly how many pixels are fully transparent, how many are partially transparent, and how many were examined.

A file with zero transparent pixels is a flat image with a PNG extension. It happens when someone exports from a tool that flattened onto white on the way out, when a JPEG was renamed, or when a screenshot of a cutout was sent instead of the cutout. It is not a subtle failure and there is no way to recover from it except to ask for the file again.

The report downloads as JSON, which sounds like a detail and is not, if you are checking sixty files and want a record of what you found.

2. Is the edge soft, or is it a stencil

This is the check almost nobody does and it is the one that separates a usable cutout from one that will look wrong everywhere.

Look at the partially transparent pixel count. A good cutout has a genuine soft transition at its boundary: a band of pixels that are neither fully opaque nor fully clear, blending the subject into whatever is behind it. That band is what makes hair look like hair and an edge look photographed rather than traced.

Now do rough arithmetic. A subject occupying most of a 2000 pixel square image has a perimeter on the order of several thousand pixels. With even a two-pixel soft edge, the partially transparent count should be in the thousands or tens of thousands.

If the report says a handful, or zero, the alpha channel is binary. Every pixel is either fully in or fully out. That cutout will look acceptable at exactly the size it was made at and jagged at every other size, it will not composite convincingly over anything, and no amount of scaling will smooth it, because there is no intermediate information to scale.

A binary mask is often a sign the cutout was made by an automatic tool with thresholding applied afterwards, or that it survived a trip through a format that could not carry soft alpha. Feathering the edge helps a little. Redoing it helps a lot.

3. Is there colour from the old background on the edge

Partially transparent pixels carry colour as well as alpha, and that colour is supposed to belong to the subject.

Very often it does not. When a background is removed, the pixels along the boundary were originally a blend of the subject and whatever was behind it, and a removal that only computes alpha leaves that blend in place. A subject cut out from a green lawn has green in its edge pixels. Put it on white and you will never see it. Put it on a dark background and there is a green rim.

Use the Colour Picker and sample a pixel whose alpha is somewhere between zero and full. Read the colour there. If it is a plausible colour for the subject, fine. If it is the colour of a background that is no longer present, you have found contamination, and you now know something useful: the fringe is in the file rather than in whatever is displaying it.

That distinction saves real time. People see a halo, assume their design tool is at fault, and re-export in three different formats before anyone checks the pixels.

4. Put it on the background you are actually going to use

Every cutout is reviewed against a checkerboard or a white page, and both of those hide things.

Add Background will composite the file against a solid colour, a gradient, or a checkerboard. Do it twice: once against near-white and once against something dark or strongly coloured. Two minutes, two images, and most defects declare themselves.

What to look for on the dark version:

• A bright or coloured rim, which is the contamination from the previous section.
• Faint see-through patches inside the subject, where the alpha is partial in a region that should be solid. These are completely invisible against white because white is what shows through.
• A hard, stair-stepped outline, which is the binary mask from section two.

What to look for on the light version:

• Leftover background the mask kept, which shows up as grey or coloured fragments floating near the subject.
• Missing pieces: a handle, a strap, a thin chair leg, anything narrow that the mask dropped.

5. Look at what is inside the file

A cutout you did not make arrived with a history, and the Metadata Remover will read it to you before you decide whether to keep it.

It reports what is physically present: EXIF including GPS and device details, XMP with editing history, ICC profiles, IPTC blocks, text chunks, and C2PA provenance. It also flags generation parameters left behind by common image generators, which is how you find out that the product photograph you were sent is a render and not a photograph.

For received assets this matters in both directions. You may be about to publish a file carrying the location it was shot at or the name of the person who made it. You may also be about to discover that an asset you were told was original is not.

Stripping it is lossless: the file is rewritten without the selected blocks and the compressed pixel data is copied through untouched. It works on JPEG, PNG, and WebP.

6. Look at it at full size

The last check is not a tool, it is a habit. Open the file at one hundred percent and look at the edge where it is hardest, which is hair, fur, fabric fringe, foliage, or anything transparent.

Thumbnails lie. A cutout that is perfect at 300 pixels and coarse at 2000 is a cutout that was made at 300 pixels and enlarged, and you will find that out on the day someone asks for a poster.

What to do with a failure

Not every failure needs the file redone. Contamination on the edge can be decontaminated in the editor. A slightly hard edge can be feathered. A small piece of leftover background can be brushed out, and a dropped strap can be restored with the brush.

What cannot be fixed downstream is missing information: a binary mask has no soft edge to recover, and a small file has no detail to enlarge. Those are the two findings worth sending back, and having the transparency report in hand makes that a factual conversation rather than an aesthetic one.

And none of it leaves your machine

Every check above runs in the browser tab. That matters here more than usual, because the files you most need to vet are frequently the ones you are least free to upload: a client's unreleased product, a supplier's catalogue under embargo, or an asset whose licence you are in the middle of deciding you do not trust.