Convert WebM to JPG

Leave the timestamp alone and you get the first frame of the file — second zero. For a camera clip that is usually fine. For a screen recording, which is what most WebM files are, second zero is a blank desktop, a browser mid-load, or the half-second before the recording actually began, because people press record and then start doing the thing.

So the timestamp control is not an advanced option, it is the main one. Type the number of seconds to the moment you care about and convert; if it is wrong, change the number and convert again. It costs a second each time because the whole operation is local, which makes trial and error a reasonable way to find the frame rather than an expensive one.

The usual way to get a picture out of a video is to play it, pause, take a screenshot and crop. That gives you the frame at whatever size the player happened to be showing it, with a scaling pass in between, and if the video was playing at half size the picture has half the detail. It also gives you whichever frame the pause landed on rather than the one you meant.

Here the frame is decoded at the video’s native resolution and encoded straight from those pixels. A 4K capture yields a 3840 by 2160 JPG whether the tab is fullscreen or a thumbnail, and the moment is the one you named rather than the one your reflexes reached.

Video compression is built on the assumption that no frame will be looked at for long. A codec spends its bits where movement is and economises everywhere else, and at thirty frames a second your eye does the averaging that hides the result. Freeze one frame and you are giving a long, still look to something that was never designed to survive one.

How bad it is depends entirely on the moment. A static shot — a slide, an unmoving interface, a talking head against a plain wall — is near the codec’s best case and comes out clean. A frame during a pan, a scroll or a cut is at its worst, because that is where the encoder had the least to work with and the most to describe. Moving the timestamp half a second is often the whole fix, and no output format is a substitute for it.

JPG earns its place by being small and by being the one image format nothing has ever refused. A full-resolution frame lands in the low hundreds of kilobytes, which is an attachment nobody complains about and an upload no form rejects. Every document editor, every email client, every listing site and every ticketing system takes one.

Its weakness is exactly the content most WebM files contain. JPG compresses by discarding fine detail near sharp transitions, which is invisible on a photograph and produces a faint halo around text, code, thin lines and interface borders. If the point of the picture is that somebody can read what is on the screen in it, take a PNG instead. If the point is to show roughly what happened, JPG is the right trade.

Browser-recorded video is essentially always 4:2:0, meaning brightness is stored for every pixel and colour for one pixel in four. It is one of the oldest tricks in image compression and it works because human vision is far more sensitive to brightness than to hue.

It shows up in exactly one place, and it is a place screen recordings are full of: coloured text and thin coloured lines on a contrasting background. Red text on dark grey, a syntax-highlighted code sample, a coloured status badge — all of them arrive slightly fringed, and the JPG then compresses that fringe rather than removing it. It is a property of the recording rather than of this conversion, and it is the reason a still from a screen capture never looks quite as crisp as the screenshot you could have taken at the time.

A single conversion gives a single frame, so a sequence means converting repeatedly with different timestamps: convert at 12, then at 31, then at 47. Each one is a couple of seconds of work and the file is downloaded before you change the number.

That is a perfectly reasonable way to pull six illustrations out of a demo recording for a set of instructions, and it is faster than it sounds because nothing is uploaded between attempts. What it is not is a way to extract hundreds of frames — for that you want a command-line tool with a frame-rate flag, and saying so is more useful than pretending a browser form is the right shape for it.

The arithmetic of this conversion is what makes uploading absurd. The input might be four hundred megabytes of screen capture; the output is two hundred kilobytes of picture. Sending the whole recording to a server so that it can send back a fraction of a per cent of it is a lot of network for a thumbnail.

Here the browser decodes the WebM with the codecs it already has, seeks to the second you asked for and encodes one frame. Nothing carrying the video moves, and the time the conversion takes is the time your machine needs to seek — which for a long recording is the only cost involved.

A fair number of WebM files are audio only — voice notes, recorded calls, clips saved out of a chat app. There is no video track in them, so there is no frame to take, and the conversion says exactly that rather than producing a blank image.

If that is the file you have, the thing you want is on the audio side: the same recording as an MP3 to listen to anywhere, or as a WAV to transcribe. A still image was never going to come out of it.