Loose 35mm film strips scattered on a light table, every numbered frame a separate picture pulled out of the take
CUTROOM

All tools / Extract Frames From Video

Extract Frames From Video.

Extract frames from a video, in the browser.

→ Grab a frameFree, no account, and your video is never uploaded

How it works.

Save a single moment as a picture you can use anywhere.

  1. Step 1

    Drag your video in

    Or click to choose a file. It stays on your device.

  2. Step 2

    Step to the frame

    Left and right move one frame, shift moves a second, space plays.

  3. Step 3

    Save the picture

    It downloads as an image you can use anywhere.

In detail

Turning a clip into a folder of stills usually means an ffmpeg command line and an argument about frame rates. Here it is a range and a cadence: mark the piece you care about, choose one frame in thirty, or one every two seconds, or a set number spread evenly across it, and the count and the total size appear before anything runs. The frames are decoded in a single pass rather than one seek each, come out at the source size with rotation applied, and arrive as one zip named by timecode. It all happens on your own machine, so a two gigabyte file never has to go anywhere.

What it gives you.

  • Pixel exact, at the decoded size

    The still comes out at the size the video decodes at, not the size of your browser window. A 4K clip gives 3840x2160 stills and a phone recording gives whatever the phone shot. Nothing is scaled unless you ask for it, and nothing is ever scaled up.

  • Frame stepping with timecode and frame number

    Left and right arrows move exactly one frame, shift plus an arrow moves a second, and space plays. The readout beneath the picture is real text carrying both the timecode and the frame number, so it reads correctly to a screen reader and copies out of the page.

  • PNG, JPEG, or WebP

    PNG is the default and keeps every pixel, because a thumbnail you are about to set type on should not start a generation down. JPEG at 92 is roughly a tenth the size, which is what a contact sheet or a training set wants, and WebP is smaller again.

  • Every Nth frame or every N seconds, into one zip

    Mark a range and take one frame in thirty, or one every two seconds, or a fixed number spread evenly across it. Up to three hundred stills at a time, decoded in a single pass rather than one seek each, and packed into one archive.

  • Orientation applied

    A phone writes its rotation as a flag on the file rather than turning the pixels, which is why sideways footage grabs sideways in most tools. The flag is read and applied here, so a portrait recording hands back portrait stills.

  • No account, nothing uploaded

    The decoding, the image encoding, and the zip all happen in this tab on your own machine. There is no queue page, no watermark, and no stranger holding a frame of footage you were only ever going to use as a thumbnail.

Straight answers.

How do I extract all the frames from a video?

Mark the range, pick a cadence, and take the zip. Extracting literally every frame of anything longer than a few seconds is almost never what people mean: ten seconds of 30 fps footage is three hundred stills and about a gigabyte of PNG, which is why this caps a single grab at three hundred and tells you the weight before you commit. Every thirtieth frame gives you one still a second, which is what a thumbnail hunt or a contact sheet actually wants. If you do need a dense run, shorten the range: the tool decodes it once, in order, rather than seeking to each frame separately, so a tight range is fast and an hour long one is refused rather than left to hang.

Why is this better than a screenshot?

Because a screenshot is a picture of a player, not a picture of your video. The player scales the frame to fit the window, so a 4K clip in a half-width browser gives you a 960 pixel wide grab. It draws its own controls over the bottom of the picture, and it may be showing a frame the compositor smoothed rather than the frame the file holds. Then the screenshot tool encodes what is on your monitor, in your monitor's colour space, usually as a second lossy generation. This decodes the frame you asked for out of the file itself and writes it straight to a PNG at the source size, so what you get is the data that was in the video and nothing else.

What size are the frames?

The size the video decodes at, with the rotation flag applied. A 1920x1080 recording gives 1920x1080 stills; a phone clip shot upright gives 1080x1920 rather than a sideways picture you have to turn yourself. If you want something smaller there are chips for 1920, 1280, and 640, which set the longest edge and let the other one follow the aspect ratio. They only ever scale down: asking for 1280 from a 640 wide source holds the source size and says so, because inventing pixels is not a resize.

How many frames can I pull at once?

Three hundred in one bulk grab, and sixty in the tray you build by hand. Three hundred is where a download stops being useful rather than where the tool gives up: three hundred 1080p PNGs is close to a gigabyte, and the readout tells you the size before you press anything. If your range asks for more than that, the range is shortened rather than the count quietly reduced, and the line under the button says exactly how much of it is covered. For a longer stretch, widen the cadence, or take it in two passes.

Which format should I pick?

PNG when the frame is going somewhere it will be edited: a thumbnail you will set type on, a still for a print, anything that will be saved again. It is lossless, so nothing is added and nothing is lost, at the cost of size. JPEG at quality 92 is about a tenth of the bytes and visually the same for a photograph, which makes it the right answer for a contact sheet, a dataset of frames, or three hundred stills you are about to skim. WebP is smaller than both and reads everywhere modern, but a few older tools still refuse it, so it is the third choice rather than the first.

Other tools