Image Metadata Glossary

Exchangeable Image File Format. A standard for storing technical metadata in image files. EXIF data includes camera make and model, lens information, exposure settings (aperture, shutter speed, ISO), GPS coordinates, timestamps, and device serial numbers. In JPEG, EXIF lives in the APP1 marker block (bytes FF E1).

Extensible Metadata Platform. Adobe's XML-based metadata format stored inside image files. XMP contains editing history, software signatures, copyright information, keywords, creator names, and AI tool identifiers. In JPEG, XMP lives in a second APP1 block beginning with the namespace string. In PNG, it lives in an iTXt chunk with description 'XML:com.adobe.xmp'.

International Press Telecommunications Council metadata format. Used primarily in editorial and news photography workflows. IPTC stores captions, credits, bylines, location names, category codes, and copyright information. In JPEG, IPTC lives in the APP13 marker block (bytes FF ED).

Coalition for Content Provenance and Authenticity. An open technical standard for embedding signed provenance metadata into media files. C2PA credentials declare the origin of an image (AI-generated, camera-captured, or edited), specify the creating tool, and include a cryptographic signature. Instagram, Facebook, LinkedIn, and Pinterest read C2PA data to apply AI labels.

JPEG Universal Metadata Box Format. The container format used to embed C2PA Content Credentials in JPEG files. JUMBF lives in the APP11 marker of a JPEG file (byte sequence FF EB). Inside JUMBF are content blocks (cb) including caBX (Content Authenticity Box), caMs (CA Manifest), and caSt (CA Store).

The user-facing name for C2PA metadata. When a platform displays 'Made with AI' or 'Created with AI tool', it is reading Content Credentials embedded in the file. Content Credentials contain a signed assertion about how the image was created.

International Color Consortium profile. Defines the color space of an image (sRGB, Adobe RGB, ProPhoto RGB, Display P3). ICC profiles ensure colors look consistent across different screens and printers. They are embedded in JPEG and PNG files. StripShot can optionally strip ICC profiles, though this may affect color accuracy when the image is viewed on calibrated displays.

A PNG file chunk type that stores plain text key-value pairs. Stable Diffusion web UIs (Automatic1111, Forge) store generation parameters including the full prompt, negative prompt, seed, and model hash in tEXt chunks with key names like 'parameters' or 'Comment'. These chunks are fully human-readable without any special tools.

An internationalized text chunk in PNG files. Supports UTF-8 encoding and optional compression. ComfyUI and other workflow-based tools store JSON workflow data in iTXt chunks. Adobe XMP metadata is also stored in an iTXt chunk with description 'XML:com.adobe.xmp'.

A compressed text chunk in PNG files. Uses zlib compression to store text data. Some AI tools use zTXt for storing compressed generation parameters. StripShot strips tEXt, iTXt, and zTXt chunks to remove all text-based metadata from PNG files.

The JPEG Application-Specific marker 1. Identified by byte sequence FF E1. Contains EXIF data (when it begins with 'Exif\x00\x00') or XMP data (when it begins with the Adobe XMP namespace string). A JPEG file can contain two APP1 blocks: one for EXIF and one for XMP.

The JPEG Application-Specific marker 11. Identified by byte sequence FF EB. Contains C2PA JUMBF data. This is the specific marker that tools check to find Content Credentials. StripShot removes APP11 blocks to strip C2PA credentials from JPEG files.

The JPEG Application-Specific marker 13. Identified by byte sequence FF ED. Contains IPTC-NAA data and Photoshop resource blocks. IPTC metadata lives here. StripShot removes APP13 blocks as part of full metadata stripping.

A hash value computed from the visual content of an image, not its exact bytes. Similar images produce similar perceptual hashes. Platforms like Pinterest use perceptual hash databases to identify known AI-generated images even after metadata removal. A pHash of 0000ffff0000ffff (example) represents a specific visual fingerprint. Unlike cryptographic hashes, small changes to pixel values produce small changes to the hash.

StripShot's mode that applies sub-perceptual pixel modifications to change an image's perceptual hash. Anti-fingerprint mode applies +/-1 RGB value changes to approximately 10% of pixels, prioritizing edge regions where changes are least visible. The result is a different pHash that does not match the original in perceptual hash databases.

The process of removing metadata by operating directly on the raw bytes of a file, without decoding or re-encoding the image. Binary-level stripping identifies metadata segments by their byte markers, removes those bytes, and adjusts the file structure. The pixel data is never modified. This contrasts with canvas re-encoding, which is lossy.

The process of loading an image into an HTML5 Canvas element, then exporting it as a new JPEG or PNG. Canvas re-encoding removes metadata because the exported file is a new image created by the browser. However, it is lossy: JPEG export applies compression, PNG conversion may occur, transparency is destroyed, and the pixel data is decoded and re-encoded, introducing generational quality loss.

The organization, led by Adobe, that developed and promotes the C2PA standard for content provenance. The CAI maintains the Content Credentials ecosystem and works with platforms, AI tool creators, and camera manufacturers to embed C2PA data in media.

The verifiable chain of custody for a digital file, tracking where it originated, who created it, and what modifications were made. C2PA Content Credentials implement digital provenance for images by embedding signed assertions at each step of creation or editing.

Data embedded in a file that describes the file's content, origin, or creation context. For images, metadata includes technical shooting data (EXIF), editorial information (IPTC), software history (XMP), provenance credentials (C2PA), and AI generation parameters (SD parameters, AI tool signatures). Metadata is stored in dedicated binary segments separate from the image pixel data.

The practice of hiding information within other data in a way that is not apparent to casual inspection. Some AI tools embed invisible watermarks in image pixels using steganographic techniques (Stable Diffusion Watermark, Midjourney Steg.AI). These are distinct from metadata and survive metadata removal. They require pixel-level analysis to detect.

An identifier embedded in an image, either visibly (a logo or text overlay) or invisibly (a pixel-level steganographic pattern). AI watermarks are typically invisible. They encode information about the generating AI tool directly into pixel values. Metadata removal does not remove pixel-level watermarks. StripShot's anti-fingerprint mode modifies the perceptual hash but is not designed as a watermark remover.