- Is my photo uploaded to a server?
- No. This tool is 100% client-side. The AI models (face-api.js TinyFaceDetector and TensorFlow.js COCO-SSD) are downloaded to your browser once, and all face detection and blurring runs locally on your device. Your photos never leave your computer. This makes the tool suitable for photos containing identifiable people, sensitive documents, or confidential scenes.
- Which file formats are supported?
- JPEG (.jpg / .jpeg), PNG (.png), and WebP (.webp) are all supported for input. All output files are saved as PNG to preserve quality after blurring. If you need JPEG output, you can re-export from any image viewer after downloading.
- What if a face is not detected automatically?
- AI face detection works best on frontal, well-lit faces and may miss profile views, obscured faces, or very small faces in a crowd. Use the manual brush tool to draw over any missed areas, then click "Apply Brush Blur" to blur those regions. You can apply the brush multiple times before saving.
- How strong is the blur?
- The tool applies a heavy Gaussian-approximation blur (radius ~20 pixels, 3 passes) that makes faces completely unrecognisable in typical photos. The blur is intentionally aggressive to ensure privacy, not just cosmetic softening. The underlying pixel data is permanently modified on the canvas, so there is no "hidden" original underneath the blur.
- Can I process many photos at once?
- Yes. You can drop or select multiple files at once. Each image is processed sequentially through the queue, and the progress is shown per file. When all are done, use the "Download All (ZIP)" button to get a single zip archive of all blurred images. Batch processing is limited only by your browser's available memory — most modern devices handle dozens of multi-megapixel photos without issue.
- Does it detect license plates too?
- Yes, partially. The tool uses TensorFlow.js COCO-SSD to detect cars, motorcycles, and buses in the image, then applies a heuristic to estimate the license plate zone in the lower-centre of each detected vehicle bounding box. This is not a dedicated plate OCR model, so accuracy varies — especially on small or angled plates. Use the manual brush to supplement auto-detection.