Face Blur — Auto Detect & Blur Faces

About this tool

Face Blur Auto detects and blurs faces in images entirely in your browser using the @vladmandic/face-api TinyFaceDetector model (~2 MB). Click 'Detect Faces' to auto-find all faces and apply Gaussian or pixelate blur. A badge shows how many faces were detected. Adjust the blur intensity (5–50 px) and switch between smooth Gaussian blur or pixelation for a CCTV-style effect. You can also draw manual blur regions by clicking and dragging on the canvas — useful for obscuring logos, license plates, or other identifying details. Download the final image as a lossless PNG.

Why use it

• AI face detection runs entirely in-browser — no image leaves your device.

• Tiny model (~2 MB) loads in seconds on any connection.

• Manual region drawing covers logos, plates, and other non-face content.

• Choose between smooth Gaussian blur or high-contrast pixelation.

• Free and instant with no signup or watermarks.

• AI face detection runs entirely in-browser via WebGL — no image leaves your device, ideal for GDPR/HIPAA workflows.

How to use

1. Upload a JPG, PNG, or WebP image containing faces.
2. Click 'Detect Faces'. The AI model (~2 MB) loads on first use.
3. Detected faces are blurred automatically and a count badge appears.
4. Adjust the blur intensity slider and switch blur style (Gaussian or Pixelate) to update live.
5. Click and drag on the image canvas to add manual blur regions for other sensitive areas.
6. Click 'Download PNG' to save the anonymized image.
7. Upload a JPG, PNG, or WebP image containing one or more faces.

When it helps

• Anonymizing faces before sharing photos on social media or in publications.
• Preparing images for GDPR-compliant use in marketing or research.
• Blurring background bystanders in event or street photography.
• Redacting identifying features from screenshots or screen recordings.
• Anonymising faces before sharing photos on social media, blogs, or in press articles.
• Preparing images for GDPR-compliant publication in marketing or research case studies.

Examples

Tips

• Run detection first, then add manual regions for anything missed — the AI handles 95% of front-facing faces but profiles or partially occluded faces sometimes slip through.
• Use Pixelate for journalistic redaction (it cannot be reversed by deblurring algorithms) and Gaussian for aesthetic anonymisation in marketing photos.
• Crank the blur radius up to 30+ px for small faces in group photos — small Gaussian radii on tiny faces can leave recognisable structure.
• If a face is detected with a tight bounding box, increase the radius or draw a slightly larger manual region around it to cover ears, jawline, and stray hair.
• Always download the PNG before closing the tab — there is no auto-save and your edits are kept only in browser memory.
• When publishing legally-sensitive redactions, prefer pixelation; certain Gaussian-blurred faces have been deanonymised by published research, while pixelation with a large block size resists those attacks.

Frequently Asked Questions

Glossary

TinyFaceDetector

A compact (~2 MB) face-detection neural network designed for real-time inference in browsers; trades a small amount of accuracy for very fast load and run times.

ONNX

Open Neural Network Exchange — a portable model format for shipping trained networks; face-api uses TensorFlow.js and a similar tensor format internally.

WebGL

A browser graphics API that exposes GPU shaders to JavaScript; face-api uses it to run the detection network on your GPU for low-latency inference.

Gaussian blur

A smoothing filter that averages each pixel with its neighbours weighted by a Gaussian curve; produces a soft, natural-looking blur.

Pixelation (mosaic)

A blur technique that replaces a region with large coloured blocks; preferred for journalistic redaction because it resists deblurring attacks better than Gaussian.

Bounding box

The rectangle returned by the detector that encloses each detected face; the tool blurs exactly that rectangle, so draw a larger manual region if hair, ears or jawline extend past it.

Browser cache

Where the face-api library and TinyFaceDetector weights are kept after the first download, so later detections in the same session start without re-downloading the model.