Touch grass. Then check what your photo gives away.

Описание и применение

Pixel Leak проверяет фотографии на наличие геолокационных метаданных и видимых текстовых признаков, затем размазывает найденные области и выдаёт чистую копию, которую можно безопасно публиковать.

Задача

Outdoor photos often reveal location details through metadata and visible text, exposing users’ home addresses, street names, and other sensitive information before they post them.

Результат

Pixel Leak identifies and removes location‑leaking metadata and pixel‑based clues from outdoor photos in about 8–11 seconds, producing a clean image that can be safely shared without revealing the user’s address or other sensitive details.

Рабочий процесс

• Upload a photo to the Pixel Leak web app
• The app reads EXIF metadata using exifr to extract GPS coordinates, capture time, and device info
• The image is processed by Florence‑2 (OCR_WITH_REGION, OD, MORE_DETAILED_CAPTION) in a Web Worker via Transformers.js
• A rules file applies regexes to model outputs to flag street names, house numbers, postal codes, phone numbers, emails, license plates, and other sensitive text
• Each flagged area is highlighted with a numbered box and assigned a severity
• The user reviews the flagged areas and chooses which ones to blur
• The app redraws the image on a canvas, applies a blur to the selected boxes, re‑encodes the image, and removes all metadata
• The cleaned image is offered for download
• The entire process runs offline after the initial 350 MB model download, with no data sent to external servers

Доказательства

Pixel Leak checks both: The metadata: GPS coordinates, capture time, device. The pixels: street and locality names, house numbers, PIN/ZIP/postcodes, phone numbers, number plates, people and screens.

Источник

A scan takes 8–11 seconds on my Mac with WebGPU.

Источник

The only network request is the one-time model download from Hugging Face (~350 MB, cached by the browser). After that you can go offline and it still works.

Источник