Start the camera · arrow through the effects · or write your own and press Apply
safe · secure · no server · works offline · fast
…or paste a file you copied in Finder, or just start typing at the cursor.
↓ scroll for about & FAQStart the camera and every frame is pulled out of the stream as a flat RGBA byte array — four bytes per pixel — passed to the JavaScript function in the editor, and painted back. The function starts as the identity transform, returning the frame untouched, so the first thing you see is your camera exactly as it is. Then you change it.
Subtracting each channel from 255 is a negative. Rotating red into green into blue is an alien colour cast. Swapping row y with row height − 1 − y turns the picture upside down. All twenty-one built-in effects are exactly that — a few lines of ordinary JavaScript, loaded into the editor where you can read and edit them, not fixed filters hidden behind a button.
There is no split between “examples” and “real” effects, because there is no difference in the code: black & white, flips, invert, sepia, threshold, posterize, pixelate, an animated wave, thermal camera, X-ray, mirror, kaleidoscope, light tunnel, squeeze, stretch, twirl, a green movie grade and an old TV set sit in one row under one word. The left and right arrow keys walk it, so you can point the camera at yourself and hold an arrow down rather than hunting for a chip.
They are here as code rather than as buttons because that is where they get interesting: thermal is a lookup into a six-stop colour ramp, squeeze and stretch are the same three lines of polar arithmetic with one constant either side of 1, pixelate averages each 24 px block in one pass and paints it back in a second, movie is three 256-entry lookup tables built once a frame instead of a curve evaluated 700,000 times, and the old TV set is a curved tube, alternate scanlines, red-green-blue phosphor stripes and a hum bar rolling down the picture — constants you can turn up until it looks like the set you remember.
Most of them turn on a single number: the block size, the cutoff, how many
levels, how far the tube bows. An effect that names it
MODIFIER gets a slider on the picture, so it
can be turned while you watch the result — full screen, on a phone,
with the camera pointed at your own face.
The slider holds no value of its own. Every move rewrites that line in
the editor and recompiles, so the code on screen is still the whole
truth about what is running: drag the dial to 8 and
const MODIFIER = 8; is what you find in the source, ready to
copy out. Declare MODIFIER and MODIFIER_RANGE in
an effect you write yourself and it gets a dial too.
Record encodes the moving picture as a clip — the canvas, at the processing size, carrying the microphone only if you ticked Sound, and never the camera track. What you see on the screen is exactly what is in the file. It opens in /mp4 BOXES, so you can read the file you just made atom by atom, and it is never written to disk unless you download it.
Record has a second mode. Timelapse keeps one frame every interval you choose — from a fifth of a second to thirty seconds — and replays them at 30 fps, so an afternoon becomes a clip you can watch: at one frame a second that is 30× faster than the room, at one frame every ten seconds, 300×. The effect is applied before anything is recorded, so a pixelated timelapse of a room is a recording that was never identifiable in the first place.
Which makes this a good place to learn: pixelate is a mean over a block, posterize is rounding to a grid, twirl is a conversion to polar coordinates and back, and each is a three-line change with your own face attached to the output. Learn maths and programming with a photo booth walks through six of them, with exercises and a 45-minute lesson outline.
The diagram above the editor is live: it reports the camera’s real
capture resolution, the milliseconds your effect spends on each
frame (with a “?” that says what that number has to
beat), and the frame rate actually reaching the screen — so when a
heavy effect slows down, you can see where the time goes. A
syntax error, a forgotten return or a wrongly sized array is
named in plain language before the video is touched, and a transform that
never returns is terminated by a watchdog instead of freezing the page.
Your camera is never uploaded or transmitted: the pixels go to a Web Worker on this origin and nowhere else, and a clip you record is held in the tab’s memory until you download it or leave.
It’s one lens on OmniViewer’s Webcam Test & Camera Capabilities: drop a file and every tab is one click away. OmniViewer opens every file format, entirely in your browser — no upload, no server.
No. OmniViewer is a static page with no server-side processing and no analytics of the video: the stream goes from your camera to the page and nowhere else. There is no endpoint to upload to. Clips you record are held in your browser tab and disappear when you leave, unless you download them yourself.
No. Nothing opens until you have granted this site camera access, and the grant is yours to make in the browser’s own prompt. What happens after that is different: once the permission exists, the Mirror and Transform tabs — the two that are nothing but a picture — open the camera on arrival rather than making you press start again every visit. The page checks whether the permission is already there without prompting for it, using the Permissions API where the browser supports the name and the presence of device labels everywhere else. Press stop and it stays stopped, and revoking the permission in your browser puts the button back.
The highest one your camera reports it can do. A camera’s real maximum is only readable from a track that is already open, so the first stream on a device is opened with no size constraint and then raised to the maximum getCapabilities() reports — one applyConstraints on the live track, no second permission prompt. The resolution picker on the Preview tab overrides it: pick a specific size, or “Camera’s choice” to send no width or height at all and see what the device volunteers on its own, which is usually a good deal less.
Because a constraint is a request, not a command. An "ideal" constraint is a preference the browser may silently ignore, and a camera already opened by another application will hand you whatever mode it is already running in. The Capabilities tab is built to show exactly this: the resolution you asked for, the range the hardware reports it supports, and the one you actually got, with the disagreement flagged.
Device labels are hidden until you grant camera access. Before the grant, a page can enumerate your devices and count them but cannot read their labels — that is a deliberate browser privacy mitigation. Press Start camera and allow access, and the names appear immediately. You can watch it happen on the Capabilities tab.
The preview and the asked/supported/got table work in any modern browser on a secure (https) page. How much the table has to show depends on the browser and the hardware: Chromium reports the widest set of camera controls, and most laptop webcams expose far fewer of them than a phone camera does. The effects need a Web Worker, and recording a clip of them needs both canvas captureStream and MediaRecorder.
Whatever the browser supports, preferring MP4 — Safari and Chrome 130+ record MP4, which opens in OmniViewer’s /mp4 toolkit with its full atom tree. Browsers that only record WebM produce a valid file that opens in the universal raw, hex and strings views, since OmniViewer has no dedicated WebM route yet. Clips are capped at 30 seconds so the recording stays in memory safely.
No, and that is half the point of recording them here. A browser capture has no camera make or model, no lens or firmware string, no serial number and no location — the capture path has no access to any of it. One line is added on the way out naming the website that made the file, and the toolkit that reads it will take even that out. Open a clip in /mp4 and compare it against a video from your camera roll to see the difference.
No. Stopping the camera stops every track, which is what turns the hardware indicator light off; loading a file, pressing close, or leaving the route all do the same. Merely dropping the video element would leave the light on, so the toolkit stops the tracks explicitly.