metadata-removal
6 cases: 1 pass, 4 partial, 1 rejected, 0 failed, 0 unsupported.
Empirical research & benchmarks
Reproducible tests of browser-based privacy and creator tools. We publish our test protocols, controlled datasets, and exact failure rates where available.
These lab pages measure EXIF, GPS, and provenance marks in files. They do not treat every app as the same behavior. One page’s comparison does not become another app’s rule. Clean the copy in this browser before sending. The file stays on this device and is not uploaded to this site.
Evaluates in-browser Tesseract.js 5.1.1 across 10 controlled multilingual document scenes (Latin, CJK, Devanagari, Arabic). German printed text passed (100% key fields); photographed scene text in Arabic and Hindi failed. Includes character error rates (CER), numeric exactness, failure taxonomy, and raw downloadable JSON/CSV datasets.
Evaluates client-side metadata sanitization across 8 controlled fixtures. Dual execution path: browser product UI for images and documents, plus engine-only regression tests for video container atoms. Documents honest boundaries: 6 PASS (4 browser product + 2 engine-only video regression), 2 PARTIAL (PDF visible page text and JPEG pixel watermark limitations), and zero cloud upload.
Captured . Counts are the cases published on each report.
Method: actual-browser-product
6 cases: 1 pass, 4 partial, 1 rejected, 0 failed, 0 unsupported.
1 cases: 1 pass, 0 partial, 0 rejected, 0 failed, 0 unsupported.
2 cases: 1 pass, 1 partial, 0 rejected, 0 failed, 0 unsupported.
Controlled checks of what metadata survives an upload, export, or AI-image download on social and creator platforms.
The index exists so a JPEG question does not get answered with a Telegram button, and a missing C2PA read does not get called a camera proof.
Integrity Policy
Direct answer: start with the file you are about to send. One page’s comparison does not become another app’s rule.
🟢 Measured on this machine · Checked November 2026 · Local JPEG and PNG samples
The 2026-11-01 page records the local viewer and cleaner. A Discord upload was not part of that run.
View protocol & test details →The 2026-11-01 page records the local viewer and cleaner. A Telegram upload was not part of that run.
View protocol & test details →See also: Labs platform methodology and Platform guides hub.
Empirical research published by Privacy labs: these pages measure files, not a universal app rule is freely citeable with attribution under CC BY 4.0.
Privacy labs: these pages measure files, not a universal app rule. (2026). Privacy labs: these pages measure files, not a universal app rule Research & Benchmark Infrastructure (Version 1.0) [Data set]. CreatorPrivacyKit. https://creatorprivacykit.com/labs/
@misc{creatorprivacykit_2026_ocr_benchmark,
title = {Privacy labs: these pages measure files, not a universal app rule Research & Benchmark Infrastructure},
author = {{Privacy labs: these pages measure files, not a universal app rule}},
year = {2026},
version = {1.0},
howpublished = {\url{https://creatorprivacykit.com/labs/}},
note = {Tested 2026-09-20. In-browser Tesseract.js 5.1.1 multilingual evaluation}
}