Convert MP4 to WebP

Nothing about the video is converted. One frame is decoded at the moment you specify and written out as an image; the rest of the file is read past and discarded, and the original is untouched.

That is worth being clear about because the extension makes it look like a video operation. It is not — there is no re-encoding, no quality decision about the video, and no output that moves. What comes back is a single picture.

A camera photograph is one exposure with all the detail the sensor captured. A video frame has been through a codec that spends its budget where the eye notices and takes it away from where it does not — and what it takes away first is detail in things that are moving.

So the same video yields very different stills. A locked-off shot of a static scene produces a frame that is genuinely sharp. A camera pan, a hand gesture or anything crossing the screen produces a frame carrying motion blur and blocking that were in the video all along and only become visible when the picture stops moving. Nothing in the conversion causes it and nothing in the conversion can remove it.

Because the codec allocates detail unevenly, the exact timestamp matters more than any setting. Half a second earlier or later — a moment when the subject is still rather than mid-movement — frequently produces a visibly better picture from the same footage.

Frames immediately after a cut are also worth trying. Video encoders usually place a full-detail keyframe at a scene change, so the first frame of a shot often carries more information than the ones that follow it. If a still is not good enough, moving the moment is the lever, not the quality slider.

A 1080p video gives a 1920 by 1080 image; 4K gives 3840 by 2160; a phone recording gives whatever it recorded. Nothing is upscaled, and nothing is reduced unless you reduce it deliberately.

This is the main advantage over taking a screenshot. A screenshot captures the video as it was displayed — scaled to the size of a window, sometimes with a play button or a progress bar across it — while this reads the frame from the file itself.

Because the destination is nearly always a web page: a video thumbnail, a poster image, a preview on a listing, an illustration in an article. WebP is meaningfully smaller than JPG at the same quality — often around a third — and every current browser reads it.

It also keeps transparency, which matters if the still will be composited with something. Where WebP is the wrong choice is anywhere outside a browser: for a document, an email attachment, a print job or another program, take JPG for a photographic frame and PNG if the frame is mostly text or graphics, such as a screen recording.

A frame from a screen recording is mostly text, interface and flat colour, and it behaves like a screenshot rather than a photograph. Lossy compression is at its worst on exactly that material — text acquires a halo and thin lines blur.

For those, take the still and keep the quality high, or use PNG instead. It is also worth checking the recording’s own resolution: a screen recording made at a scaled-down size can never produce a sharp still, and the fix is to record at full resolution rather than to work on the frame afterwards.

Video generators return short MP4s, and a still from one is wanted constantly: a thumbnail for the post, a reference frame to feed back into an image prompt, a preview for a client who is not going to watch a five-second clip. Screenshotting the player gives you the player as well, at whatever size the window happened to be.

A frame grab takes the pixels the encoder wrote, at the clip’s own resolution, with no interface around it. The one thing to expect is softness on motion — generated video carries the same per-frame compression as any other, so a frame from a fast pan is blurrier than the clip looks in motion. Pick a still moment and the grab is as sharp as the source allows.

The usual production job is not one frame but several — a poster image, a hover preview, a set of chapter thumbnails. Taking them from consistent moments rather than by eye makes the result look deliberate: the first frame after each cut, or a fixed offset into each section.

Keep the video, obviously, and keep a note of the timestamps used. Regenerating thumbnails at a new size or in a new format is then a repeat of a known operation rather than a hunt for the frame somebody liked last year.

Decoding runs in the browser tab on your own processor. Nothing is uploaded, so a large video does not have to be sent anywhere before a single frame can be taken — which is the practical benefit as much as the privacy one, since uploading a two-gigabyte file to retrieve a 40-kilobyte image is a poor bargain.

There is no account, no queue and no daily allowance. The free tier takes files up to 100 MB, which a few minutes of 4K passes, and a hundred files to a batch.