Something wrong with this tool?

About Face Blur Photo Online

Upload a photo and the tool obscures every face it finds using on-device face detection — no third-party AI service ever sees your image. Detection runs with Google's BlazeFace model loaded locally on the server, and the actual blurring is done with libvips/Sharp. The output is a high-quality JPEG with the original aspect ratio preserved, so it drops straight into reports, social media, or news articles without further editing.

Three obscuration styles are available. Gaussian blur applies a smooth, recognisable-shape-but-not-identifiable haze and looks natural in editorial photography. Pixelate replaces each face with a coarse colour mosaic — slightly more jarring but irreversible against modern depixelation attacks when the block size is large enough. Solid black box is the most assertive option and is the right choice for whistleblower photos, court evidence, or any image where any partial visibility is unacceptable.

The Face padding slider extends the obscured area outward from each detected face by a configurable fraction (0 = tight crop on facial features, 1 = doubled bounding box). Increase padding if hair, ears, or chin are still visible after detection. If no face is detected the tool returns the original image unchanged with a note — try a clearer, front-facing photo or another image.

How to use this tool

How to blur faces in a photo

  1. Upload your photo

    Drop a JPEG, PNG, WebP, GIF, or HEIC file onto the upload area, or tap to browse. Up to the per-tool size limit shown next to the field; oversized files are rejected before any processing.

  2. Pick a blur style

    «Gaussian blur (soft)» keeps the face shape but unrecognizable. «Pixelate (mosaic)» replaces it with a coarse grid. «Solid black box» is the most aggressive — use it for whistleblower photos or court evidence.

  3. Set strength and padding

    Strength = blur radius (Gaussian) or pixel size (pixelate). Higher numbers obscure more. Face padding extends the obscured area outward; bump it up if hair or chin still show after the first run.

  4. Press Blur faces

    The server runs face detection with a local BlazeFace model (no external API), then composites the obscured regions onto the original. A JPEG with the original aspect ratio is returned, plus the count of faces found.

  5. Download and verify

    Open the result card, download the JPEG, and zoom to 100% to confirm every face is covered. If one was missed, increase padding or strength and run again. If detection failed entirely the original photo is returned unchanged with a note.