Pull still images out of a video. Scrub to a moment and grab that frame, or capture one every few seconds and download the set as a ZIP. Your browser does the decoding, so anything it can play, this can read — and nothing is uploaded. Two things this gets right that catch most frame grabbers out: it waits for the frame to actually arrive before drawing, rather than capturing whatever was on screen a moment earlier, and it reports when the browser seeks to a different time than you asked for, which it does whenever the nearest keyframe is elsewhere.
| Format | Size | Use for |
|---|---|---|
| PNG | Largest | Editing, cropping, transparency |
| JPEG | Smallest | Thumbnails and contact sheets |
| WebP | Small | Web use, smaller than JPEG at equal quality |
Whatever your browser can play — typically MP4 and MOV with H.264, and WebM. If a file will not load, it is a codec your browser has no decoder for rather than a limit here. There is no upload and no conversion step.
Because browsers seek to the nearest decodable frame, and in a heavily compressed video those keyframes can be seconds apart. The tool tells you the time it actually landed on rather than pretending it hit yours exactly.
PNG if you are going to edit or crop the frame — it is lossless but large. JPEG for a thumbnail, where the size difference is enormous and the quality loss invisible. WebP is smaller than JPEG at the same quality and is supported almost everywhere now.
Not efficiently — that needs a real decoder loop rather than a seek per frame, and a minute of 30fps video is 1,800 seeks. Interval capture is what this is for. For every frame, use ffmpeg on your machine.
The video's own, by default. You can scale down while keeping the aspect ratio; scaling up is not offered because it adds nothing the source did not have.
So they sort correctly. Without the padding, frame-10 comes before frame-2 in every file manager and inside the zip, which is the opposite of useful when you wanted to step through them.
No. It is read as a local blob and decoded by your browser — nothing is sent anywhere, which matters for footage you would not upload.
A `<video>` element does the decoding, which means whatever your browser can play this can read — MP4 and MOV with H.264, WebM, and whatever else the platform supports — with no WASM demuxer and no upload. Capturing is a seek followed by a canvas draw, and the honest part is that a seek does not necessarily land where it was asked. Browsers seek to the nearest decodable frame, and in a long-GOP encode those keyframes can be seconds apart, so a request for 3.00 seconds can return 2.50. Rather than hide that, the tool reports the time it actually landed on whenever it differs by more than a frame, because someone comparing the timecode burned into their footage against what they typed deserves to know which one is true. The capture waits on the `seeked` event and then, where the browser has it, on `requestVideoFrameCallback` — without that second wait the canvas is drawn from whatever was on screen before the seek completed, which produces a set of frames that are each one step behind the time in their filename. Interval capture has a specific off-by-one worth naming: seeking to exactly the reported duration lands past the last decodable frame in every browser, and what comes back is either the previous frame repeated or a blank canvas, so an "every two seconds" pass over a ten-second clip silently ends on a duplicate. The final capture is therefore pulled a frame's worth inside the duration. Filenames are zero-padded to the width of the total, because `frame-10` sorts before `frame-2` in every file manager and every zip listing — which is precisely the wrong order when the point was to step through them. Output is PNG, JPEG or WebP, optionally scaled down while keeping the aspect ratio, and a batch is zipped. Nothing is uploaded: the video is read as a local blob URL and never transmitted.