Nothing leaves the machine
The file is decoded by the browser, drawn into a canvas with a source rectangle offset, and re-encoded. All three steps are local. No request is made when you pick a file and none is made when the crop appears, which you can see for yourself in the network tab, or prove by disconnecting and cropping anyway.
Crop, not resize
The largest rectangle matching the target ratio is selected and everything outside it is discarded. No interpolation happens, so the pixels that survive are the pixels that were there — a crop cannot soften an image the way a resize can. The consequence is that the output dimensions are whatever the arithmetic gives: a 4032 x 3024 phone photo cropped to 1:1 comes out 3024 x 3024, not 1080 x 1080. Ratio and pixel size are separate requirements and they need two separate operations.
Ratios worth knowing
| Use | Ratio | Common pixel size |
|---|---|---|
| Square feed post, profile picture | 1:1 | 1080 x 1080 |
| Tall feed post | 4:5 | 1080 x 1350 |
| Stories, reels, shorts | 9:16 | 1080 x 1920 |
| Video thumbnail, widescreen still | 16:9 | 1280 x 720 or 1920 x 1080 |
| Link preview card (og:image) | 1.91:1 — type 191 and 100 | 1200 x 630 |
| Wide profile header | 3:1 | 1500 x 500 |
| Pinboard image | 2:3 | 1000 x 1500 |
| 4x6 photo print | 3:2 | 1800 x 1200 at 300 ppi |
| A4 page | 1:1.414 — type 1000 and 1414 | 2480 x 3508 at 300 ppi |
Profile pictures are masked to a circle almost everywhere, so anything within roughly 15% of the edge of a 1:1 crop is at risk. Keep the face well inside.
Choosing the alignment
Centre is right often enough to be the default and wrong often enough to be worth a second look. Portraits usually want top, because heads sit above the middle of the frame and a centre crop takes the forehead off. Landscapes want top if the sky is the subject and bottom if the foreground is. Cropping a wide shot to vertical is where left and right earn their keep, and the report tells you exactly how many pixels came off each side so you can reason about it rather than guess.
How much you lose, before you commit
The area kept when going between ratios is the smaller of the two ratios divided by the larger. Going 16:9 to 9:16 keeps about 32% — you throw away two thirds of the picture. Going 4:3 to 1:1 keeps 75%. Going 3:2 to 4:5 keeps 53%. If those numbers are alarming for the image in question, the honest conclusion is that the source was framed for a different shape and a fresh shot or a padded layout will serve better than an aggressive cut. The tool will happily perform the aggressive cut; it just tells you what it cost.
Questions people ask
Can I drag the crop box where I want it?
No — the choices are centre and the four edges. That covers the large majority of ratio crops without needing a canvas interaction that behaves badly on a phone. Anything requiring a specific off-centre framing is editor work, and the built-in Photos app on either desktop platform will do it in less time than describing the position would take.
The output is not the pixel size the platform asked for.
Correct, and deliberate. This matches the ratio and leaves the resolution alone, because scaling and cropping are separate decisions and combining them hides which one softened the image. Crop first, then resize the result. Doing it in that order also means the resize has the maximum number of pixels to work from.
Does cropping reduce quality?
The surviving pixels are untouched, so the geometry costs nothing. The re-encode is where any loss happens: a JPEG in and a JPEG out has been through the transform a second time. Choose PNG for a lossless round trip, or leave the quality at 90 or above if the file has to stay JPEG. A PNG or WebP source held in the same format loses nothing at all.
What happens to an animated GIF?
Only the first frame survives. A canvas holds one still image, so the output is that frame as a single picture regardless of what the format was. Animation editing is outside what a canvas can do and needs a tool that understands frame timing.