Blur anything out of a video

Pixelate faces, license plates, screens and names — in a tool that physically cannot leak your footage, because it never leaves your device.

Drop a video here
MP4 or MOV · censored on your device, not on a server
Choose video
On-device No upload Irreversible censoring Free

The one video you'd never upload is the one you need to blur

Think about what's actually in a video that needs censoring: your child's face, a stranger who didn't consent to being filmed, your home address on a parcel, a license plate, a screen recording with your email open in another tab. Sending that footage to a stranger's server in order to make it private has always been a contradiction — you'd be sharing it more widely than if you'd done nothing.

Zimblu resolves the contradiction by not having a server in the loop at all. Your browser decodes the video, censors the region you choose, and re-encodes it — all on your own hardware. There is no upload progress bar on this page because there is no upload. If you're skeptical (you should be, it's a strong claim), load the page, switch off your Wi-Fi, and watch the tool keep working.

Made for the awkward frames

Faces

Bystanders, kids, anyone who didn't agree to be in your upload. Tap Auto-detect or drag a box per face — the model runs entirely on your device.

License plates

Dashcam clips and driveway videos travel further than you think once posted.

Screens & documents

Screen recordings with notifications, addresses on parcels, badges and paperwork caught mid-pan.

Locations

House numbers, street signs, storefronts — anything that pins the footage to a place.

Three steps, one honest re-encode

  1. Drop the video. It opens instantly — the file stays where it already is.
  2. Place a region. Drag the box over what should disappear; pull the corner to resize; stack several boxes if you need them. Pick pixelate (crisp mosaic) or blur (frosted), and set the strength. For faces, the Auto-detect button runs a small on-device model on the current frame and seeds the boxes for you.
  3. Export. One re-encode at a bitrate matched to your source. In our test runs a 30-second 1080p clip took about 4 seconds on a recent laptop. Audio is copied through untouched.

A technical note you won't get from most tools: weakly gaussian-blurred text and faces can sometimes be partially recovered. That's why both modes here work by resampling — the region is collapsed to a handful of pixels and rebuilt, so the original detail is discarded, not veiled. The censoring is one-way by construction.

Blurring, answered

Is it safe to blur a sensitive video here?

Safer than anywhere that asks you to upload. Zimblu does the censoring inside your browser, on your own hardware — the video never reaches a server, ours or anyone else's. You can disconnect from the internet after the page loads and the tool keeps working, which you're welcome to try.

Can the blur be reversed to reveal what's underneath?

No. Zimblu doesn't paint a soft filter over the region — it resamples the area down to a handful of pixels before scaling it back up, so the detail is physically discarded, not hidden. That matters: research has shown that weak gaussian blurs and swirl effects can sometimes be partially undone. Throwing the pixels away is the safe approach, and it's what both the pixelate and blur modes do here.

Does the blur follow a moving face?

In the browser tool the region stays where you place it for the whole clip, so make it generously larger than the subject's movement. For a face that walks across the frame you'd add a wider box, or trim the clip to the seconds that matter. Automatic motion tracking is the kind of heavier job the Zimblu iOS app is built for.

Does it find faces automatically?

Yes. There's an Auto-detect button that runs a small face-detection model on the current frame and adds a generously-padded blur box over every face it finds. The model is downloaded only the first time you click — about 3 MB, cached after that — and runs entirely on your device, so even the detection step never sees a server. Detection is per-frame in this browser tool, so for a face that moves a lot you'll want to either resize the box or also use the manual region path; per-clip face tracking is a Zimblu iOS app feature.

How long does exporting take?

Censoring requires re-encoding the video once, at a bitrate matched to your source. In our test runs, a 30-second 1080p clip exports in about 4 seconds on a recent laptop — roughly seven times faster than the video's own length. Audio is copied across untouched.

While you're here