Crop an image
Drag a rectangle over the part you want to keep.
How it works
A crop discards everything outside the rectangle. The pixels inside are left alone, copied row by row out of the decoded bitmap, unchanged. What is not free is writing the result back out. If you ask for JPEG, those surviving pixels are put through a second generation of DCT quantisation on top of the compression they already carry, and edges that were already softened soften again.
Dedicated lossless JPEG croppers avoid this by rewriting the compressed data directly, so they can only cut on 8- or 16-pixel MCU boundaries and cannot honour an arbitrary rectangle. This tool decodes to pixels and re-encodes, so choose PNG when you need the remaining pixels to come out byte-identical, and JPEG or WEBP when the file size matters more.
Transparency and animation drop out
JPEG carries a second cost: the format has no alpha channel, so transparency inside the crop is composited onto white before encoding, and the result note says so when that happened. One further consequence of decoding: an animated GIF gives up its first frame, and the crop is a still image.
Why the rectangle uses percentages
The four coordinates are percentages of the source, not pixels, because a rectangle in pixels is only meaningful for one resolution. The default is 10 / 10 / 80 / 80, a centred crop that removes the outer tenth. It means the same framing on a 4000×3000 camera original and on the 800×600 copy you exported for the web.
Pixel coordinates do not travel: crop at 400,300 on the original, halve the image, and the same numbers now point at the middle of the frame instead of a tenth of the way in. Percentages resolve to whole pixels once, at the moment of the crop, by rounding against the actual decoded dimensions, so the result panel reports the rectangle it produced, in the form `4000×3000 → 3200×2400`, and you can read the real pixel size rather than infer it.
Ratios the platforms actually want
Most platform requirements are ratios, and cropping is how you get the ratio; scaling to the exact pixel count afterwards is what resize-image is for. Square 1:1 covers profile photos and Instagram's original feed post. 4:5 is the tallest Instagram will show in the feed, delivered as 1080×1350. Stories and Reels, TikTok and YouTube Shorts all want 9:16, or 1080×1920. YouTube thumbnails and most in-stream video are 16:9 at 1280×720.
Open Graph and Twitter summary cards want roughly 1.91:1, in practice 1200×630, which is the one people most often get wrong because a square image posted as a link preview gets its top and bottom cut off. That cut is the reason to crop deliberately: every platform crops for you when the ratio is wrong, almost always from the centre, and the centre of a photograph is rarely where the subject's face is.
A crop is not redaction
Cropping is a geometric operation and does not, in itself, remove metadata. In most editors the EXIF block survives intact, still holding GPS coordinates, camera serial number, lens and the original timestamp. Removing a person or a street sign from the frame does not remove where and when the photograph was taken from the file. Treat a crop as composition, never as redaction.
The same block holds the orientation flag, which is a note to the viewer to rotate the picture rather than a property of the pixels, so an upright-looking phone photo may be stored sideways. A dedicated EXIF viewer and metadata stripper is planned for the privacy cluster and is not built yet; until it ships, if metadata matters, check the file with something that reads EXIF rather than assuming any step in this pipeline handled it.
Questions
Is my image uploaded to a server?
No. Open your browser's developer tools, switch to the Network tab, and crop a file. You will see requests for code and WebAssembly codecs, and nothing that carries your image. The crop happens in a Web Worker on your machine, which is also why there is no queue and no daily limit.
Why are the coordinates percentages instead of pixels?
So they stay correct across resolutions. A percentage rectangle means the same framing on a camera original and on a downscaled export; a pixel rectangle is tied to the one size it was measured on. The result note reports the pixel rectangle the percentages resolved to, so you can still see exactly what you got.
Does cropping reduce the quality of what is left?
Not in itself — the pixels inside the rectangle are copied unchanged. Encoding them again can. PNG output is exact. JPEG and WEBP re-compress, at quality 90 here, which is visually close but a real second generation of loss. If you are cropping a JPEG and want to keep editing it, take PNG out of this tool and compress once at the end.
Can I crop to an exact pixel size like 1080×1080?
Crop to the ratio here, then send the result to resize-image, which takes exact dimensions and appears with your bytes already loaded. Doing it in that order is deliberate: cropping to a ratio and then scaling gives you the size you asked for without stretching the subject, which is what happens if you force both at once.