An independent maintainer toolbox

What this project is

CreatorPrivacyKit is maintained independently. It is not a company, a certification body, or a lab. Product questions go to @mollycoddle330 on X. Source: github.com/ggggghhhhyyyy/creator-privacy-kit.

The flagship tool is Metadata Cleaner on the homepage: inspect and strip supported C2PA, EXIF, GPS, XMP and container metadata from JPG, PNG, WEBP, MP4 and MOV in this browser tab.

The rest of the kit covers jobs creators actually hit: AI image provenance signals, GIF to Live Photo / Motion Photo encoding, local video compression, local OCR (English, Chinese, Japanese, Korean, Spanish, French, German, Portuguese, Hindi, Arabic), a local PDF toolkit, an on-device text assistant, a remote AI Text Detector.

Local-first, with named exceptions

“Local-first” is an architecture choice. File bytes for Metadata Cleaner, Local OCR, Local PDF Toolkit, Video Compressor, and GIF to Live Photo are processed in the browser. They are not uploaded to a CreatorPrivacyKit processing API.

Named exceptions:

• AI Text Detector — remote analysis. Submitted writing goes through the CreatorPrivacyKit Worker to TypeSafe System One. If TypeSafe times out, rate-limits, or returns an invalid response, the Worker falls back to OrcaVerify Text 1.0. This tool is not local-only; the text leaves the device.
• AI Image Origin Detector — local inspection stays in the tab. Optional “Verify with OpenAI” sends the selected image through the CreatorPrivacyKit Worker to OpenAI’s content provenance API (C2PA / SynthID). Fetching a remote image URL also uses the Worker, transiently, without storing the file.
• Private AI Text Assistant — Privacy Scan runs without a model download. Optional MiniCPM5-2B inference runs in the browser after the user chooses to fetch the model weights.

If a page says “no upload”, that claim applies to that tool’s default path. It does not mean every button on the site is offline.

What we will not claim

Cleaning metadata does not remove invisible pixel-domain watermarks such as SynthID.

A missing C2PA or SynthID signal is not proof that an image is human-made.

We do not promise Instagram, X, TikTok, or search-ranking outcomes. Platforms can relabel, recompress, or ignore file-level credentials.

HEIC/HEIF iPhone originals are not a supported Metadata Cleaner input today. Export JPEG/PNG/WEBP first, or see the methodology notes.

GIF to Live Photo encoding and file-structure checks are browser-tested. Apple Photos, Samsung Gallery, and Android import are not yet verified on a physical device.

Video Compressor prefers hardware encoding when the browser exposes it. That is not a guarantee that hardware was used, and a size preset is a target, not a promise.

Local OCR ships 11 language packs (English, Simplified Chinese, Traditional Chinese, Japanese, Korean, Spanish, French, German, Portuguese, Hindi, Arabic). It does not ship a Vietnamese recognition pack. A translated UI is not an OCR language pack. Auto is English, not script detection. For empirical accuracy metrics across languages and document types, see our multilingual OCR benchmark.

We do not certify other people’s files, publish user counts, or claim an accreditation that does not exist.

Editorial principles

Tool pages describe the current implementation, including limits.

Articles prefer primary sources: official platform docs, standards text, and measurements we can reproduce in a browser. Community reports stay labeled as such.

When a platform limit or a tool behavior changes, we update the relevant page and its dateModified. CSS, header, or spelling tweaks do not by themselves bump article dates.

If a factual claim is wrong, we fix that page. For a material error, we note the change on the same page.

Public identity

Maintainer contact: @mollycoddle330 on X.

Source repository: github.com/ggggghhhhyyyy/creator-privacy-kit.

How claims are tested: Methodology.

Empirical research & benchmarks: Laboratory and OCR Benchmark.

Contact

Contact: hello@creatorprivacykit.com