Blur faces online — automatic single and batch face blurring
Detect faces automatically and blur, pixelate or cover them — in one photo or a batch of up to 100. The detector, its runtime and the model all run inside your browser. Your images never leave your device.
Runs in your browser — your input is not uploaded to ToolsNow.
Proportional to each face
JPEG and WebP only
Your files are processed locally in your browser and are not uploaded to ToolsNow.
Your images never leave your device. The detector, its WebAssembly runtime and the model are served from ToolsNow and run inside your browser; no image, filename, dimension or detection result is sent anywhere. The model locates faces only — it does not identify, recognise or compare anyone, and nothing is stored. Downloaded files are freshly encoded from a canvas, so EXIF, GPS coordinates and embedded thumbnails from the original are not carried over. Automatic detection may miss small, obscured, angled or partially visible faces. Always inspect the completed image before publishing or sharing it. Blur and pixelation transform pixels rather than discarding them; solid cover is the only option that removes the underlying detail outright.
How to use
- Add an image, or several. Single mode gives you a full editor. Batch mode processes a folder of photos one at a time. JPEG, PNG and WebP.
- Wait for detection. The face-detection model downloads once per visit — about 12 MB — then runs on your device. Nothing is uploaded at any point.
- Review every region. This is the step that matters. Detected faces are outlined; weak detections are marked “needs review”. Add anything the detector missed by dragging on the image.
- Choose the effect. Blur is the default. Pixelate gives a mosaic. Solid cover is the only one that removes the underlying detail outright.
- Check the padding. The detector’s box is tight around the eyes, nose and mouth. Padding grows it proportionally so the forehead, jaw and ears are covered too.
- Download. One image, or a ZIP of the whole batch. Files are re-encoded, so the original EXIF and GPS data are not carried over.
Questions
Are my photos uploaded anywhere?
No. The face-detection model and its WebAssembly runtime are served from ToolsNow as ordinary static files, then everything runs inside your browser tab. No image, filename, dimension or detection result is sent to any server — ours or anyone else’s. You can confirm this yourself: open your browser’s network panel, process an image, and you will see no request carrying it. After the model has loaded you can disconnect from the internet entirely and the tool keeps working.
Does this recognise or identify people?
No. The model locates faces — it returns rectangles and a confidence score, nothing more. It does not identify, match, compare or classify anyone, no face data is stored, and there is no database for it to be compared against.
Will it find every face?
No, and you should not assume it has. Automatic detection reliably misses small background faces, faces at sharp angles, faces partly hidden by hair, hands, masks or sunglasses, and faces in poor light. In our own testing on a crowd photograph the detector found only a handful of the faces present. That is why the tool shows you every region before you download and lets you draw your own — always inspect the finished image before publishing or sharing it.
Which effect should I use?
Solid cover if the stakes are real. Blur and pixelation transform the original pixels rather than discarding them, so information survives in a reduced form; published research has shown that pixelated and blurred content can sometimes be reconstructed. Blur at a strong setting is fine for casual sharing, and solid cover is the honest choice when the consequence of identification is serious.
Why is there padding around each face?
A face detector’s box is tight around the eyes, nose and mouth. Blurring exactly that leaves the forehead, jaw, ears, hairline and often part of the chin visible, which is frequently enough to recognise someone. Padding is proportional to each detected face rather than a fixed number of pixels, so the same setting behaves identically on a small thumbnail and a 40-megapixel photograph.
How many images can I process at once?
Up to 100 files per batch, each up to 25 MB and up to about 40 megapixels. They are processed strictly one at a time — decoding a hundred full-resolution photographs simultaneously would exhaust the tab’s memory, since a decoded image costs four bytes per pixel regardless of how well it compressed. You can cancel at any point and keep whatever has already finished.
What happens to my photo’s metadata?
It is removed, as a consequence of how the export works: the result is drawn onto a canvas and encoded afresh, and none of the original file’s EXIF survives that. GPS coordinates, camera model, timestamps and embedded thumbnails are all gone. This is not the same as anonymity — the visible content is still the visible content.
Which formats are supported?
JPEG, PNG and WebP, in and out. Those are the three that decode reliably across every browser this site supports. HEIC, TIFF, AVIF and SVG are refused with a clear message rather than accepted and then failing — and SVG is excluded intentionally: it is a document format that can carry script.
Does the interface freeze on large batches?
It should not. Detection is synchronous and takes tens of milliseconds per image, so it runs in a Web Worker rather than on the page’s main thread. The page stays scrollable and the cancel button stays responsive throughout, which is also what makes cancellation work rather than merely queueing.