Face Blur Tool: Pixelate or Blur Faces in a Photo
The tool finds faces with Google's MediaPipe face detector and covers each detected face with a mosaic or a Gaussian blur. Check the result before you share it: any face the model misses stays sharp.
Drop your photo here or click to select
Supports JPG, PNG, and WebP - up to 50 MB
Runs in your browser: your files and text are not uploaded. Free, no account needed.
Last updated: · Reviewed by the QuickTooly Team
Detection and its limits
How Faces Are Detected
Detection uses the MediaPipe Tasks Vision face detector with the BlazeFace short-range model. On the first run your browser downloads the model from storage.googleapis.com and the WebAssembly runtime from cdn.jsdelivr.net. A face is covered only when the model is at least 50% confident.
The short-range model is designed for faces close to the camera, such as selfies and portraits. MediaPipe describes it as best for faces within about 2 metres.
Faces the model can miss
- Small faces: people in the background of a group, crowd or street photo.
- Faces in profile or turned away from the camera.
- Partly covered faces: masks, sunglasses, hands, hair, or another person in front.
- Dark or blurred faces: poor light and motion blur lower the confidence score.
The result line says how many faces were covered. Compare it with the number of people in the photo. You cannot add or move a box by hand, so cover a missed face in an image editor or crop it out.
Pixelate, Blur and the Strength Slider
- Pixelate (default): the face box is shrunk by the block size and scaled back up without smoothing, which gives square blocks. The slider sets the block size from 5 to 40 px; the default is 15.
- Gaussian Blur: the face area plus a margin is blurred with the slider value as the radius, then clipped to the face box. Browsers that cannot blur on a canvas, such as older Safari, pixelate instead and say so under the result.
Both settings are in image pixels, not screen pixels. In a high-resolution photo a face can be hundreds of pixels wide, so 15 px blocks still leave a detailed mosaic. Raise the slider until the face is unrecognisable at full size, not only in the small preview.
Only the detection box is covered. Hair, ears and the edge of the chin can fall outside it.
The download is always a PNG at the original dimensions, ending in "-faces-pixelated" or "-faces-blurred". Metadata from the original file is not copied into it.
Face Blur Questions
Why did the tool miss a face?
The BlazeFace short-range model is tuned for faces close to the camera. Small background faces, faces in profile, and faces partly hidden by masks, glasses or hands often fall below its 50% confidence threshold. Cover missed faces in an image editor.
Can I select a face manually?
No. The tool only covers the faces the model detects. There is no way to draw or adjust a box.
Is a pixelated face impossible to recognise?
Not always. With blocks that are small compared to the face, the shape and colours can still identify a person, especially to someone who knows them. Use a larger block size or blur radius and check the result at full size.
Why is the download larger than my original photo?
The output is always PNG, which is lossless. A photo saved as PNG is usually much bigger than the same photo as JPEG. Convert the result to JPG if file size matters.
After Blurring
- PNG to JPG Converter: turn the PNG output into a smaller JPG for sharing.
- EXIF Viewer & Remover: check what other copies of the original photo reveal, such as GPS location.