Photo Scrubber — On-Device Face Blurring & Metadata Stripping
Photo Scrubber is an experimental local tool using MediaPipe and BlazeFace to blur faces and strip metadata for public photo privacy.
Key Takeaways
- Photo Scrubber is an experimental local privacy tool designed to blur faces and remove metadata from photos taken in public or protest settings.
- It combines GPT-6 Astra, Google MediaPipe compiled to WebAssembly, and the BlazeFace model to detect and blur likely face regions without sending images to a remote service.
- The tool addresses two main risks: accidentally exposing identifiable bystanders and leaking metadata such as time, camera settings, or location.
- It is useful for photographers and event documenters who want a faster way to reduce identity exposure before sharing images.
- Because it is experimental and may miss faces or leave metadata fields untouched, users should test it carefully and still manually review sensitive photos before publishing.
Photo Scrubber Adds Local Face Blurring for Protest Photos
Meta description: Photo Scrubber is an experimental local tool using GPT-6 Astra, MediaPipe, WebAssembly and BlazeFace to blur faces and remove photo metadata.
Tags: photography, tools, privacy, face detection, WebAssembly
Featured image: Abstract friendly illustration of a camera, a blurred rectangle over a face-like shape, and a shielded photo file, with no real people, no logos, and no text.
Photo Scrubber local face blur and metadata removal describes an experimental tool for people who photograph public gatherings. The project author says they created it after taking a photograph of some protesters and deciding not to share images that kept strangers identifiable. It is aimed at photographers, event documenters, and ordinary users who want a simpler way to reduce identity exposure before publishing. The source says the tool uses GPT-6 Astra, Google's MediaPipe C++ library compiled to WebAssembly via @mediapipe/tasks-vision, and the BlazeFace face detection model.
Why this tool exists
The starting point was a practical privacy problem. The author took a photograph of some protesters, then thought about not liking to share photographs of strangers with identifiable faces. That concern is common in public and protest photography, where a frame can contain more than the intended subject. It may include bystanders, children, people who are not there as organizers, or individuals who do not want their image associated with a movement, location, or event.
Photo files can also carry metadata. In many camera and phone images, metadata can include details such as time, camera settings, and sometimes location. If a photograph is published without review, that information can reveal when and where it was taken, or even a route taken during a day of shooting. The tool's stated purpose is to address two visible risks at once: blur faces and remove metadata.
How it works
The tool is described as local, which means its design points toward processing images on the user's own environment rather than sending them to a remote service. The source does not provide a full technical report, but the named components are recognizable. Google's MediaPipe C++ library is used for machine learning pipelines. WebAssembly is a technology that allows compiled code to run inside web pages and other sandboxed environments. The package @mediapipe/tasks-vision is the JavaScript-facing layer for MediaPipe vision tasks. BlazeFace is a face detection model, meaning it finds likely face regions rather than identifying who those faces belong to.
The author also says they had GPT-6 Astra build the experimental tool. The word experimental matters. The source does not state accuracy numbers, supported file formats, platform requirements, update policy, or whether metadata removal handles every field in every camera file. Those details are important for a tool that may be used in sensitive situations. Readers should treat it as a privacy aid, not as a complete safety system.
What readers should expect
For photographers and documenters, the value is speed and friction reduction. If faces can be blurred and metadata removed before upload, it becomes easier to share images without exposing identifiable strangers by accident. That can help in event reporting, community documentation, and personal posting where the subject agreed to be photographed but nearby people did not.
Still, readers should check the project documentation, test the tool with non-sensitive sample images, and review the output before publishing. Face detection can miss small faces, faces turned away from the camera, partial faces, or people in low light. Metadata cleaning can be incomplete if the tool does not handle all fields used by a specific camera or platform. If a photo contains highly sensitive content, the safest approach is still manual review and conservative sharing.
Photo Scrubber is most interesting because it connects a common creative workflow with a practical privacy need: taking images in public spaces while trying not to expose people who did not choose to appear.

Comments (0)