Metadata & AI Privacy Glossary
An authoritative technical reference on digital image metadata standards, C2PA Content Credentials, AI generation tags, and digital forensics terms.
An open technical standard established by Adobe, Microsoft, Intel, and the BBC that embeds cryptographically verifiable digital provenance manifests directly into media files. C2PA records who created an asset, which AI generator or camera was used, and every subsequent editing pass.
How MetaRinse AI handles it: Strips JUMBF (JSON Universal Metadata Box Format) containers from JPEG, PNG, and WebP bitstreams, severing the external provenance validation chain without altering visual pixel contents.
Standardized under ISO/IEC 19566-5, JUMBF is a container architecture used to encapsulate rich, nested binary data inside digital files. It is the primary transport mechanism for C2PA Content Credentials inside JPEG APP11 markers and PNG standalone chunks.
Key insight: Standard image viewers ignore JUMBF boxes, but social media platforms (Instagram, TikTok) scan them upon upload to auto-apply "AI-generated" disclosure badges.
A specification for image file formats used by digital cameras and smartphones. It stores camera make/model, lens parameters, focal length, exposure time, ISO speed, date/time stamps, and precise GPS latitude/longitude coordinates.
Privacy concern: Geotagged EXIF data frequently leaks personal home addresses, daily travel routes, and device fingerprints to anyone who downloads the unstripped photo.
The global standard for news and press photo metadata. IPTC records copyright notices, creator credit lines, licensing terms, captions, and keywords. Many modern AI editing tools write "AI-generated image" directly into IPTC DigitalSourceType fields.
How MetaRinse AI handles it: Excises IPTC IIM records and Adobe Photoshop 0xFFED APP13 markers completely.
An XML-based metadata standard created by Adobe that can be embedded into JPG, PNG, PDF, and TIFF files. It stores extensive edit histories, Lightroom development presets, mask layers, and software identifiers.
Risk factor: XMP often retains deleted author names, original filenames, and internal server paths even after an image has been cropped or resized.
A technology developed by Google DeepMind that embeds an imperceptible digital watermark directly into the pixel values of AI-generated images, audio, and text. Because SynthID alters actual pixel values rather than file headers, it cannot be stripped simply by removing metadata headers.
Important distinction: MetaRinse AI strips all container-level metadata (EXIF, C2PA, PNG chunks), but clearly discloses that pixel-baked watermarks like SynthID require lossy geometric transformations or pixel resampling.
Unlike cryptographic hashing (where 1 flipped bit completely scrambles the hash), perceptual hashing produces fingerprints based on human visual features. Platforms like Pinterest, Google Images, and Etsy use perceptual hashes to identify duplicate images and track republished assets.
MetaRinse AI Innovation: Applies sub-perceptual micro-variations to pixel distributions that shift the perceptual hash (e.g. Hamming distance > 10) while preserving 100% human-visible image fidelity.
A proprietary section within EXIF data used by manufacturers like Canon, Nikon, Sony, and Apple to record hardware-specific diagnostic info, including camera body serial numbers, total shutter actuations, and internal sensor temperatures.
Forensic tracking: Camera serial numbers in MakerNotes allow investigators and automated crawlers to link seemingly unrelated photos back to a single physical device.
An architectural computing model where all computational workloads (decoding, memory parsing, byte stripping, hash computation, and canvas rendering) execute strictly within the user's local web browser via JavaScript and WebAssembly, without transmitting any data over the internet.
MetaRinse AI Guarantee: Your files never touch a server, ensuring zero data retention and zero exposure to third-party data breaches.
