Blur faces in a video automatically
Filming a school play, a street scene, a protest, an office walkthrough or a delivery on your doorstep usually means recording people who never agreed to appear online. This face blur video tool finds the faces in every part of your clip, follows them as they move, and hides them with a strong blur, pixel blocks, a solid patch or an emoji — so you can share the video without exposing anyone.
Detection uses Google’s MediaPipe face detector with the BlazeFace short-range model (Apache-2.0), running inside your browser. Before anything is saved you can scrub through the video and see every face box, tap a box to keep one person visible, and draw extra boxes for faces the detector missed, each with its own time range. The video is never uploaded.
How to use the Face Blur Video
- Drop your video on the upload area. Its duration, resolution, frame rate, codecs and size are shown.
- Choose how often to look for faces (every frame is most accurate, every 5th is fastest) and the sensitivity, then press Find faces. Progress and speed are shown as it scans.
- Scrub the preview slider or click the timeline above it: blue bars mark frames where faces will be hidden, grey bars faces you chose to keep visible, amber frames where no face was found. Tap a face box to keep that person visible; drag on the preview to add a box for a missed face and set its time range in the list.
- Pick the effect — Blur, Pixelate, Emoji or Solid — the shape, strength and padding around each face.
- Press Blur faces, check the result with the Before/After buttons, and press Download video.
Face Blur Video features
Detection plus tracking
Faces found on sampled frames are linked into tracks and their boxes interpolated for the frames in between, so the blur moves smoothly with each person.
Review before saving
A timeline shows where faces were found, and the preview shows every box at any moment of the video.
Keep or add faces
Switch a tracked face off to keep it visible throughout, or draw manual boxes with start and end times for anything missed.
Four ways to hide
Strong blur, chunky pixelation, an emoji over each face with pixelation underneath, or a solid patch.
Small-face scan
Square tiles along the frame are always checked; the optional small-face mode adds a finer grid for crowds and distant people.
Private and offline
Detection, editing and encoding run on your device. The model only locates faces — it does not recognise who anyone is.
How faces are found and followed
Each sampled frame is shrunk and passed to BlazeFace short-range, a fast detector designed for faces within a few metres of the camera that looks at a 128 × 128 pixel version of the image. The whole frame is checked, then square tiles across it, so a face that is small in a wide 16:9 frame still fills enough of the model’s view. Overlapping hits are merged, and every detection keeps its confidence score, so changing the sensitivity re-uses the scan instead of repeating it.
Detections are then linked from one sampled frame to the next by overlap and distance into tracks. For frames that were not checked, the box is interpolated between the neighbouring detections, and a face that disappears for a moment keeps its box for a short gap. Finally Mediabunny (MPL-2.0) re-encodes every frame with the faces hidden, using your browser’s WebCodecs encoder, while the soundtrack is copied unchanged.
What automatic face blur can miss
No detector is perfect. Faces that are tiny, far away, seen in profile or from behind, partly covered by hands, masks, hats or hair, blurred by fast motion, in strong shadow or turned upside down may be found late, lost for a few frames or not found at all. Reflections in windows and mirrors, faces on posters, TVs and phone screens, and other identifying details such as name badges, school logos, tattoos and number plates are not looked for. Always scrub through the result before sharing, raise the sensitivity to “Catch more”, look for faces in every frame, and add manual boxes where needed. For plates, screens and signs, use Blur Video.
Tips
- A padding of 30–50% also covers hair and ears, which can identify people too.
- Blur at high strength or Solid is safest; very light blur can leave a face recognisable.
- Long videos take a while: shorten them first with Trim Video, or check every 3rd frame.
- For still photos, Face Blur does the same job for single images and batches of photos.
Face Blur Video FAQ
How do I blur faces in a video automatically?
Open the video here and press “Find faces”. Scrub through to check the boxes, tap any face you want to keep visible, add boxes for faces that were missed, then press “Blur faces” and download the result.
Does the blur follow people as they move?
Yes. Faces are detected on sampled frames, linked into tracks and interpolated in between, so the blurred area moves with each face. Check fast movements, where a box can lag slightly — a larger padding helps.
Why was a face not blurred?
Small, turned, covered or motion-blurred faces are hard for any detector. Try “Catch more” sensitivity, look in every frame, tick “Also look for small faces”, or draw a box over the face by hand for the time it is visible.
Can I keep my own face visible?
Yes. Tap your face box on the preview — it turns dashed and that person stays visible for the whole video. Tap again to hide it.
Can a face blur be reversed?
Not from the saved file: the pixels are replaced. Very light blur or small pixel blocks can leave a face recognisable, so keep the strength high, or choose Solid.
Is my video uploaded or used for face recognition?
No. Everything runs in your browser. The detector only finds where faces are; it does not identify anyone, and the video never leaves your device.