Skip to content
iSpeedIMG

Blur faces in photos

Protect bystanders, friends and anyone who didn't sign up to be published. Faces are found by an on-device model, blurred right where they are — and you draw the regions the AI missed. Nothing is ever uploaded.

How to blur a face

  1. 1. Drop your photo into the tool. A small face detection model (~230 KB, one-time download) scans it locally and marks every face it finds.
  2. 2. The preview blurs each detected box. Untick any face that should stay visible in the review list.
  3. 3. Press Draw region and drag over anything the model missed — then pick PNG or WebP and hit Download.

The AI, honestly

This tool runs YuNet — a compact face detector published by the OpenCV project and licensed Apache-2.0, the same license family that protects the compiler-tool code you already rely on. It runs on-device through WebGPU or WebAssembly; the 230 KB model is served from this very site.

Good at

  • Clear, upright faces in decent light — portraits and street shots alike.
  • Groups and multiple faces in one photo.
  • Tiny downloads: the whole detector is smaller than a photo of a cat.

Weak at

  • Steep angle shots, hands over faces, backs of heads.
  • Very small faces — zoomed-out crowd shots are the classic miss.
  • Guaranteed coverage. Always, always review before you publish — that's what the drawn regions are for.

Output behavior

  • The output keeps the original dimensions. Everything outside your boxes is untouched pixel-for-pixel.
  • Blur strength scales with each region: big faces and small faces get proportionally the same relative coverage.
  • Pixelate mode replaces the smear with hard mosaic blocks — some platforms crack down on recognizability differently, so pick the one you prefer.
  • Output metadata: like every tool here, the exported file is brand-new with no EXIF or GPS attached.

Privacy

The photo is read into memory, scanned by the model, blurred on a canvas and re-encoded — all locally. No upload, no processing queue, no server log to worry about. Close the tab and everything is discarded.

Browser & device compatibility

  • Runs in every modern browser via Web Workers; WebGPU makes it faster, WebAssembly is the automatic fallback.
  • Unlike heavier AI tools, the face model is ~230 KB — no big first-use download.
  • Very large photos are downscaled internally for detection, then blurred at full resolution afterwards.

Troubleshooting

  • "That file type isn't supported" — the file isn't a JPG, PNG, WebP or AVIF image. Rename or re-export first.
  • "Your browser couldn't decode this file" — the name is recognized but this browser can't read it (often AVIF on an older browser), or the file is corrupted.
  • A face is still visible — raise the blur strength, switch to Pixelate, or draw a tighter region over exactly that spot.
  • Blur looks weak on a huge face — the strength adapts to region size; bump the slider and the preview updates instantly.

Related tools

  • Blur License Plate — Blur license plates automatically, then review and draw any extra regions.
  • Remove EXIF — Strip EXIF, GPS location and hidden metadata from photos — lossless and private, on-device.
  • Compress Image — Reduce JPG, PNG and WebP file size with quality you control.

Frequently asked questions

How do my faces get detected?

A neural face detector (YuNet, open source under Apache-2.0) runs inside your browser and returns a location box for every face it recognizes — the classic pipeline behind countless camera apps. Blurring happens in a second step from those boxes.

Are my photos uploaded anywhere?

No. Detection and blurring both run on your device; the photo never leaves it. The face model is tiny (~230 KB) and is served from this same site — no third-party CDN, no telemetry.

Can a blurred face still be recognized?

Strong blur makes identification practically impossible, and you control the blur strength. Check the preview: if any feature is still readable, raise the blur strength or switch to Pixelate — the mosaic blocks are more aggressive at the same setting.

Can I blur the background of a photo instead of a face?

Yes. Drag any region over the background and it is blurred or pixelated exactly like the face boxes — the AI suggests faces automatically, and anything else such as a background, a whiteboard or an address plate you draw by hand.

What kinds of faces does it miss?

The detector is trained on clearly visible, roughly upright faces. It can miss very small, strongly tilted, covered or motion-blurred faces. That is why you can draw extra regions by hand — always give the result a full review before sharing.

Which format should I download?

PNG keeps the image exactly as blurred (lossless) and is the safe default. WebP is noticeably smaller with no visible difference. JPEG is smallest but re-encodes lossily — usually still perfectly fine for social-posting photos.