GPT-6 Astra builds a tool that blurs faces in photos
In a short post dated 29 September 2026, the author describes a small experimental tool that identifies faces in photographs and blurs them out automatically. The motive was personal. The author took a photograph of some protesters, then thought about how they don't like sharing photographs of strangers with identifiable faces.
The author says they had GPT-6 Astra build the tool, so the code was written by the model at their request.
The technical stack is named in two lines. The tool uses Google's MediaPipe C++ library, compiled to WebAssembly via @mediapipe/tasks-vision, plus the BlazeFace face detection model. In other words, detection rests on existing Google components rather than anything new, and the model's job was to assemble them into a working face-blurring utility.
The post is brief. It gives no accuracy figures, no performance numbers and no test results, and it does not describe the protest itself.
Key facts
- The post is dated 29 September 2026 and describes an experimental tool that identifies faces in photographs and blurs them automatically.
- The author wanted it after photographing some protesters, because they don't like sharing photographs of strangers with identifiable faces.
- The author says GPT-6 Astra built the tool at their request.
- It uses Google's MediaPipe C++ library, compiled to WebAssembly via @mediapipe/tasks-vision, plus the BlazeFace face detection model.
Why it matters
The story is a small, concrete example of a person turning a privacy worry into a working tool by asking a model to build it. The worry is specific: sharing photos of strangers whose faces can be identified. The tool answers it by blurring those faces automatically instead of leaving it to manual editing.
Who it affects
Anyone who photographs people in public and then shares the pictures, as the author did after photographing protesters. The post frames the concern around the author's own discomfort with sharing identifiable strangers, not around a wider audience.
How to use it
The post does not give a way to try the tool. No URL or link to the tool itself, no code repository and no license are given in the source text. What a reader can take from it is the recipe: Google's MediaPipe C++ library compiled to WebAssembly via @mediapipe/tasks-vision, together with the BlazeFace face detection model.
How solid is it
This is a first-person account from the author of the tool, and it calls the tool experimental. No accuracy figures, performance numbers or test results for the face detection or blurring are given. The components named, MediaPipe and BlazeFace, are Google's; the post itself is a few sentences long.
Risks and caveats
Because the tool is described as experimental and no accuracy results are given, there is no evidence in the post of how reliably it finds every face. A missed face would stay identifiable. The post also gives no details of the prompt, the number of iterations or the time it took GPT-6 Astra to build the tool.
“I took a photograph of some protesters, then thought about how I don't like sharing photographs of strangers with identifiable faces.”
— Author of the post, 29 September 2026