Blur the background of a video

Portrait-mode blur for a video you already recorded: the person stays sharp, the room behind them goes soft.

Blur everything behind the person in an MP4, MOV or WebM, like portrait mode for video. It is free, needs no account and adds no watermark, and your files are not uploaded: everything runs in your browser on your device. The first run downloads the person finder and its engine (about 28 MB, or about 12 MB with the Fast option) to your browser, and after that the tool works offline.

Drop videos to blur the background ofor click to choose files. Batches welcome.

    Made for people. Fast is rougher around hair and in dim light.

    Can run slower than the video plays; the time left shows as it runs.

    Working…

    Done

    Your files are processed on this device and never sent to a server. Reload the page to clear everything from memory.

    Phones blur the background of photos, and video calls blur it live, but a clip you have already recorded keeps every detail of the room: the unmade bed, the whiteboard with client names, the family photos on the shelf. This page finds the person in each frame and blurs everything else, so the recording can be shared without the room.

    A worked example: a 90-second 1080p selfie video for a job application, filmed in a kitchen. Drop it here; Blur is already chosen at 6, which on a 1080-pixel-tall frame is a blur of about 26 pixels, enough to make text and faces in the background unreadable while the shapes stay natural. Each frame is examined and redrawn, the sound is copied, and you get application-blurred-background.mp4. For a job portal or LinkedIn, run it through Compress video next if the file is too large to upload.

    Push the slider to 10 to hide more, or down to 3 for a gentle depth-of-field look. Anything that is not a person, such as a pet on your lap or a product you hold up, counts as background and is blurred too.

    How it works

    1. Drop the video; Blur is already chosen.
    2. Set how strong the blur should be.
    3. Press Blur background and wait while each frame is redrawn.

    In your browser, not on a server

    Mediabunny (MPL-2.0) decodes the video in your tab. Each frame is scaled down and given to a Google MediaPipe selfie segmenter (Apache-2.0) running in WebAssembly: by default the multiclass one, a 16 MB model that rates every pixel as background, hair, skin or clothing, or, if you choose Fast, the binary one, 0.25 MB, that only separates person from background. The person’s outline is softened at the edge and averaged with the previous frame so it does not flicker, the background is blurred or replaced with your colour or picture, and your browser’s own encoder (WebCodecs) writes the new MP4, with the sound copied across.

    Compared with upload-based tools: Online video background removers upload your recording, cap free exports at a few seconds or add a watermark, and many keep the file. Here the video never leaves your device, whatever is in the room behind you.

    Verify it yourself. Give the page a short while after your first run to save itself for offline use, then turn off Wi-Fi or switch to airplane mode and run the tool again. Or open your browser's developer tools (F12, or Cmd-Option-I on a Mac), pick the Network tab, and watch it while you convert. The panel above does the same count for you.

    Questions

    Will the blur hide text and faces behind me?

    At the default strength, text on a wall or screen and faces in photos behind you become unreadable. Check the result before sharing: if you move quickly, a sliver of background can stay sharp next to you for a frame or two.

    Does it work with more than one person?

    Yes. Everyone the model sees as a person stays sharp, whether one or several.

    Why does something I hold get blurred?

    The model looks for people, not objects. A mug, a phone or a pet counts as background unless it overlaps your body.

    Is the video uploaded?

    No. The whole job runs in this tab, on your device.

    How long does it take?

    Every frame is examined and redrawn on your device, so it can take longer than the video plays. The default finder, Better edges, runs on the graphics chip where the browser offers one. On a MacBook Pro with an M3 Max chip a 10-second 30 fps clip took about 6 seconds in Chrome and about 10 seconds in Safari, so a 90-second clip would take about 1 to 1.5 minutes (we timed 10 seconds, not the longer clip). Where the browser cannot use a graphics chip (software rendering, some virtual machines and browser settings) it falls back to the processor, and the same clip took about 55 seconds, so a 90-second clip would take about 8 minutes. The Fast option runs on the processor and took about 5 seconds in Chrome and 8 to 9 seconds in Safari on the same clip, so it is only a little quicker than Better edges on a graphics chip and several times quicker than Better edges without one, but its edges are rougher around hair and in dim light. Phones vary and we have not benchmarked one with the graphics chip: a single run on an iPhone in Safari, before this change, took roughly 5 to 10 minutes for a 2-minute video, which is a rough guide, not a measurement. Slower devices take longer, and so do longer or 60 fps videos. The page shows the time left once it is running; keep the tab open while it works.

    Why does the browser console mention a Google address?

    The open-source MediaPipe engine that finds the person tries to send its own usage log to a Google address. This site’s security policy makes the browser refuse that request, so it fails and you see an error in the console. Your video is not part of it and it never leaves your device.

    More video and audio tools

    Other conversions

    Guides