Crop video by drawing the rectangle you want to keep

Crop video by saying which pixels stay

Every other free video cropper on the first page of results asks for the same thing first: the file. It goes up to a server, waits in a queue, gets cut, and comes back. That is a reasonable way to build a website and a strange way to treat a recording of somebody’s living room. This page does the cut in the tab you already have open. The video is read off the disk in pieces, every frame is drawn and re-encoded here, and the finished MP4 is written to storage that belongs to this browser. Nothing is uploaded, there is no account, and there is no queue to be at the back of.

The interaction is the one you expect. Drag a rectangle over the picture, pull its corners or edges until it holds what you want, and press the button. What is less expected is that the four numbers beside it are not a fallback for people who could not manage the mouse — they are the same rectangle by another route, in the source picture’s own pixels, and dragging simply writes into them. Typing 0, 0, 720, 720 and dragging to the same place produce the identical file, because both go through the same arithmetic before either one reaches the encoder.

Cropping is not the same job as changing the aspect ratio

These two get confused constantly, and the difference decides which tool you actually want. Cropping starts from the rectangle: you name the part of the picture that survives, and the resulting shape is whatever falls out of that. Changing the aspect ratio starts from the shape: you say 9:16, and the tool works out the rectangle for you, centered, because centered is the only guess it can defend.

That guess is right surprisingly often and catastrophic the rest of the time. A speaker standing on the left of a wide frame is centered out of their own video; a caption burned along the bottom edge is cut in half; a logo in a corner disappears. When the pixels matter, the rectangle is the only one of the two that can promise anything. When they do not — when a platform simply demands a square and any square will do — the aspect ratio changer is fewer decisions and will also offer to pad with a color or a blurred copy of the video instead of cutting, which this page deliberately does not do. Padding adds picture that was never recorded, and a tool called a cropper should not quietly do that.

The shape buttons here are a bridge between the two. Pressing 1:1 or 9:16 writes the largest rectangle of that shape that fits, centered on whatever you were already looking at rather than on the frame. Then you keep dragging. They are a starting point, not a mode, and nothing locks afterwards.

Why this tool draws every frame itself

The media engine this site uses has a crop option built in, and using it would have been three lines. It was measured on 10 September 2026 against both browser engines the tools are tested in, and in WebKit — Safari, and every browser on iOS — it does nothing at all. Not an error, not a warning: the output file is written, it has exactly the dimensions that were requested, and it contains the entire original picture squeezed into that shape.

The cause sits one layer below the engine. Drawing a frame of decoded video onto a canvas can take a source rectangle — the part of the source to copy — and WebKit ignores that rectangle when the source is a decoded video frame, scaling the whole thing instead. The same silent substitution happens when the crop is expressed as a bitmap crop rather than a draw, which is worth saying because that is the workaround usually recommended; it was measured too, and it fails identically. Expressing the crop as a narrower visible region on the frame crashed the tab outright.

What does work, in both engines, is to draw the whole frame exactly once into a canvas that is already the size of your rectangle, positioned so the parts you did not select fall outside the canvas edge. No source rectangle exists anywhere in that operation, so there is nothing for an engine to ignore. That is what runs here, on every frame. It is slightly more work per frame than asking the engine nicely, and it is the only version of this feature that is true on an iPhone.

The reason this matters beyond one bug is what a test would have caught. A test that checks the output dimensions passes in both engines, because both write the promised dimensions. Only a test that decodes the finished file and reads its pixels can tell the difference — so that is what guards this page: it crops the left half of a known clip and the right half of the same clip, and requires the two results to contain different colors, in Chromium and in WebKit.

The MP4 this tool writes

MP4

Full name	MPEG-4 Part 14
Extension	.mp4, .m4v
Format type	Video container — usually H.264 or H.265 with AAC
MIME type	video/mp4

Even edges, and what the re-encode changes

Every number this page shows is even, including the left and top offsets, not just the width and height. H.264 codes in pairs of pixels: an odd width is rounded up by the encoder whether or not the interface admits it, and an odd left offset shifts the color grid half a step and softens the edge for no reason anybody asked for. Rounding here, before the number is ever displayed, means the encoder has nothing left to round and the rectangle on screen is the rectangle in the file.

Removing part of a frame means painting that frame again, which is why a second H.264 pass is unavoidable here. No lossless route exists; a cropper advertising one has quietly stopped doing one of the two things. The soundtrack fares better — an AAC stream rides through untouched and anything else is converted so the container stays playable — while chapters, subtitles, timed metadata and any additional tracks are dropped on the way. Resolution is left alone: your rectangle’s pixels come out at the size they went in, never resampled to something rounder.

Afterwards the written file is reopened and interrogated. Its picture size is pulled back out of the container and printed on the receipt next to the rectangle you asked for, and when those two numbers disagree the receipt says so in as many words. A page that instead echoed the request would be making a claim nobody could ever catch it getting wrong, which is not much of a claim.

What to run before or after this

A crop is seldom the whole job. Shortening the clip in the video trimmer first leaves fewer frames for this page to repaint, which makes the crop quicker and spares footage you were discarding anyway from a pointless generation of loss. Should the result then run into an upload limit, the video compressor works backward from a byte ceiling and weighs what it wrote. If a single picture is really what you were after — a thumbnail of the region you just selected — the frame extractor lifts one out at full resolution, while video to GIF is the truthful answer for a destination that wants a loop instead of a video.

Doing the work here still spends something: your processor and, more importantly, your memory. A long or high-resolution source can exhaust a tab, and an exhausted tab simply restarts itself with no explanation whatsoever — so the warning about size arrives before the encode does, never after. Nothing is turned away for being big; it is your machine to spend. Pressing cancel halts the encode and deletes the half-written scratch file, and any strays left behind by an earlier visit are swept up a day later. Rather than take this paragraph on faith, watch it yourself: the network verification guide shows how.