Extract a frame from a video

How to use it

Load the video (or drop it above), scrub near the moment you care about, then use the four fine controls: one frame or one tenth of a second, backwards and forwards. The ← and → arrow keys do the same. When the picture is right, hit «Save this frame»: the downloaded file is named after the video plus the timecode, for example video_00-01-23-450.jpg, so you always know where it came from.

Why this beats a screenshot

A screenshot captures the video at the size you happen to be watching it: if the player is 600 pixels wide, that is the picture you get. Here the frame is redrawn at the real resolution of the file (videoWidth × videoHeight), so a 4K video gives you a 3840 × 2160 image with no player controls, no cursor and no screen glare. Pick JPG for a light file, PNG for no loss at all, handy for plates, signs and anything you need to read.

Frame rate: why we ask for it

Browsers never tell a page how many frames per second a video runs at, so the «one frame» step is simply 1/fps: 25 fps for European (PAL) broadcast, 24 for cinema, 30 or 60 for most phone footage. If you do not know, press «Measure from video»: the tool plays less than a second muted, measures the gap between the frames actually shown and, when it lands within 2%, snaps the value to the standard ones (23.976 · 24 · 25 · 29.97 · 30 · 48 · 50 · 59.94 · 60 · 90 · 100 · 120). It is an estimate: with variable frame rate footage the number is only indicative, and 23.976 and 24 are 0.1% apart, so no measurement of this kind can tell them apart for sure.

How exact is «one frame», really

Let us be straight about it: a web page has no «go to frame 1372» command. It can only ask for a point in time, and this tool asks for the middle of the frame worked out from the frame rate you gave it, t = (n + 0.5) / fps, so rounding cannot drop you back on the frame you were already on. Hence the real limits: if the frame rate is wrong, or the footage is variable frame rate, a step can land on the neighbouring frame; and with interframe codecs (H.264, HEVC, VP9, which is nearly everything) the picture has to be rebuilt from the last key frame, and browsers are not equally precise about it, Safari in particular may stop at the closest frame it can display. That is why the timecode you see in the result box and in the file name is not the one requested but the real one the browser reports after the jump: the image you save and the timecode next to it are always the same frame. If you need certainty about frame number N, for evidence, say, use an editor or ffmpeg: no web page can guarantee it.

The grid, for picking a cover

Four, nine or sixteen frames taken at regular intervals show you the whole video at a glance, the fastest way to choose a thumbnail or find the right scene. Each frame is taken in the middle of its slice, so the first one is not the usual black frame, and every thumbnail is a button: tap it and the tool goes back to that exact time and saves the image at full resolution. This needs the browser to know how long the video is; with some WEBM files and interrupted recordings it does not, so the grid stays off while single frames keep working. If what you actually need is to show the video to an AI chat (ChatGPT, Claude), there is a dedicated tool: Video for AI chats, which prints the time under every frame and prepares the text to paste too.

If the video will not open, or comes out black

Browsers decode what they decode: H.264 and VP9 nearly always, HEVC/H.265 often not (which covers a lot of iPhone and action cam footage), and neither many MKV files, old AVIs nor ProRes. When that happens you get a clear warning, and the fix is to convert the file to MP4 H.264 first with Convert video. Vertical phone clips come out with the same orientation you see in the player; if your browser misreads the rotation metadata, the «Rotate» menu fixes the picture before saving. Everything runs on your device: the video is never uploaded to any server.