Detect faces locally, review every numbered region, add anything the detector missed, and export a privacy-ready image or batch. The detector, editor and exporter run on this device; your image pixels are never sent to a server.
All detections start selected. Click a box or checkbox to exclude it, move or resize any region, and add a manual region when the detector misses something.
Drop photos here or click to select
JPEG, PNG, WebP, GIF, BMP, TIFF, HEIC or AVIF when supported by your browser
Drag a box to move it. Drag a handle to resize. Drag empty space to pan; use the wheel or zoom buttons.
Face detection is a useful first pass, not proof that every person has been found. Small, rotated, covered or partially visible faces can be missed, and a non-face can occasionally be marked. Review the full image before export, especially when anonymity matters.
Your browser downloads a face-detection model, then runs it in a background Web Worker on this device. Image pixels move only between memory owned by this tab. There is no upload endpoint, cloud inference or account history, and MediaPipe telemetry is blocked inside the detector worker.
The model and processing runtime are hosted with the site and cached after first use. Once those assets have loaded, disconnecting the network does not stop detection, editing or export.
Strong pixelation with face-and-hair coverage is the default because a light blur may preserve recognizable structure. Opaque shapes and overlays are stronger still. “Estimated head” is deliberately named: it expands the box geometrically and is not hair segmentation.
Exports remove metadata by default, including GPS coordinates, camera identifiers and embedded thumbnails. Compatible metadata can be retained for same-format JPEG, PNG and WebP exports, but doing so can reintroduce identifying information and always requires an explicit opt-in.
No. The detector is downloaded as a static model and runs in a Web Worker inside your browser. Your image pixels, filenames, detections and edited regions are not sent to LocalConverter, Google or another processing service.
Yes. Tiny, rotated, obscured and partially visible faces are difficult for every automatic detector. Thorough mode checks tiles and rotations, but you should still inspect the whole image and add manual regions where needed.
Strong pixelation or a fully opaque overlay generally removes more recognizable structure than a subtle blur. The safest choice depends on the image, so the tool previews every effect and asks you to review the result before export.
Yes by default. Canvas export discards metadata and the tool does not copy it back unless you explicitly enable compatible metadata preservation. That option may retain GPS, camera and author information, so leave it off for privacy-focused sharing.
Yes. Add a batch, let the browser scan each file sequentially, review flagged images, apply one effect configuration and export the results as a ZIP. Processing and ZIP creation both happen locally.