How to tell if a photo is AI-generated from its metadata (C2PA, XMP, EXIF)
The article explains how to detect AI-generated images by examining metadata standards such as C2PA, XMP, and EXIF. It details specific metadata fields and tools that can reveal whether a photo was created or modified by generative AI, providing a technical guide for verification.
Background
- C2PA (Coalition for Content Provenance and Authenticity) is an open technical standard that cryptographically signs a photo's origin, editing history, and device info — think of it as a "nutrition label" for digital media, backed by Adobe, Microsoft, Sony, and others.
- XMP (Extensible Metadata Platform) is Adobe's metadata schema, embedded in file formats like JPEG or PNG, that stores editing history, software used, and timestamps.
- EXIF (Exchangeable Image File Format) is the decades-old standard embedded by cameras and phones recording shutter speed, GPS location, make/model, and other capture details.
- AI-generated images often lack normal EXIF camera fields (no lens, aperture, GPS) or reveal the generator tool name (e.g., "Midjourney," "DALL-E") in XMP. C2PA signatures may be absent or show a creation tool like "Adobe Firefly."
- The article walks through practical forensic checks (using free tools like ExifTool, PhotoME, or online C2PA viewers) to examine these metadata layers before trusting a photo's authenticity.