Cropping is the one image operation that is genuinely simple: you keep a rectangle of pixels and discard the rest. No arithmetic is applied to the pixels you keep. Which makes it all the more irritating that cropping tools so often deliver something other than the rectangle you selected.
There are three usual reasons — the preview is not the crop, the photo's orientation is being handled differently by the tool and by your eyes, or the tool re-encodes the result harder than it needed to. All three are avoidable, and knowing about them tells you what to check in whatever tool you use.
The preview and the crop must be the same rectangle
The commonest defect in crop tools is that the rectangle drawn on screen and the rectangle actually cut are computed by two different pieces of code. They agree most of the time and diverge at the edges: at the image boundary, at very small selections, at unusual aspect ratios, or after a zoom. The result is a crop that is a few pixels off from what you framed, which is invisible on a landscape and fatal on a document scan or an ID photo.
The structural fix is for the preview overlay and the crop to be computed by the same function, so there is only one rectangle in the system and no way for two answers to disagree. Filum does that, and shows the precise pixel values in a readout under the preview so you can confirm the numbers before committing rather than inspecting the download afterwards.
Why phone photos crop wrong
A photo taken with a phone held sideways is usually stored the way the sensor read it, with a separate EXIF orientation tag saying "rotate this 90 degrees before showing it." Every normal viewer applies the tag, so you see the picture upright. A tool that reads the raw pixels without applying the tag sees it on its side.
When that happens, you drag a rectangle over an upright preview and the tool cuts the corresponding rectangle out of a sideways image. The crop is wrong by a rotation, and the failure looks bizarre enough that people assume they made a mistake. The fix is to apply the orientation before the selection is offered, so the image you frame and the image being cut are the same one. Filum does that, so the region you drag is the region you get.
What cropping costs, and the case where it should cost nothing
The pixels inside the rectangle are copied exactly — cropping never resamples or blurs them. But JPG and WebP have to be re-encoded to be saved, and that re-encode is a second lossy pass over pixels that were already lossy. It is usually imperceptible; it is not free, and it is worth knowing it happened. Filum tunes that encode against a measured perceptual score and shows the number with the result. PNG is stored losslessly, every pixel identical.
There is one case where the right answer is to do nothing. If you select the entire image, there is no crop — and a tool that re-encodes it anyway has degraded your file to perform an operation with no effect. Filum returns your original file back unchanged and tells you nothing was cropped. That sounds like a small courtesy, but a no-op path that quietly damages the input is one of the more common ways image tools lose quality for no reason at all.
Where the crop happens
Cropping is frequently a privacy operation. People crop to remove a face from the edge of a photo, to cut a name and address off a document scan, to trim the surrounding desk out of a picture of a form. A tool that uploads the file to crop it has received the uncropped original — the exact thing you were trying not to share — before doing anything about it.
Filum's Crop Image runs the crop inside your browser on your own device, so the full image is never transmitted. It supports JPG, PNG, and WebP, and refuses formats it cannot handle immediately rather than failing silently partway through.