Glossary
AI image provenance glossary
Thirty terms you meet when checking whether an image is AI-generated, each with a one-sentence definition, the detail that matters, and exactly how isGenAI uses it in a report.

test-signed.pngPNG file · controlled example
- Signature + IHDR
- iTXt XML:com.adobe.xmp
- tEXt parameters
- tEXt prompt / workflow
- tEXt invokeai_metadata
- eXIf
- caBX C2PA manifestrecord
- IDAT pixels
- IEND
Content Credentials found
A signed manifest describes how the file was made.
Signer, recording tool, actions and the signature check result are reported separately, so you can weigh each one.
Open the Content Credentials checker →Terms
C2PA
C2PA is the open standard for Content Credentials: signed manifests that record how a file was made.
Content Credentials
Content Credentials are the consumer name for C2PA manifests: signed records of how an image or video was made.
Provenance
Provenance is the recorded history of where a file came from and what changed it. Learn the kinds of evidence and how isGenAI reads each.
SynthID
SynthID is Google DeepMind’s invisible watermark for AI-generated images, audio, video and text.
IPTC Digital Source Type
IPTC DigitalSourceType is the metadata field that declares how an image was made, including trainedAlgorithmicMedia for AI.
EXIF
EXIF is the camera metadata standard in JPEG, TIFF, WebP and PNG. Learn what it records, what AI tools write into it and how isGenAI reads it.
XMP
XMP is the XML metadata packet in JPEG, PNG, WebP, TIFF and PDF files. Learn what it carries, including the IPTC AI source type.
JUMBF
JUMBF (ISO/IEC 19566-5) is the box format C2PA uses to store manifests inside JPEG, PNG, WebP, HEIF, TIFF, MP4 and PDF.
AI watermark
An AI watermark is a visible mark or an invisible signal added to generated media to show its origin.
Deepfake
A deepfake is AI-made or manipulated media that falsely appears authentic. Read the EU AI Act definition, the Article 50(4) duty and the file checks.
AI-generated image
An AI-generated image is a picture produced by a generative model. Learn how generators record that fact and how isGenAI reads the evidence.
AI image detector
An AI image detector decides whether a picture was AI-generated. Learn the two approaches, pixel classifiers and evidence readers.
Prompt metadata
Prompt metadata is the generation prompt and settings some AI tools save inside the image file.
PNG text chunk
PNG text chunks (tEXt, zTXt, iTXt) are keyword-value records inside a PNG where AI tools save prompts and XMP.
ComfyUI workflow
A ComfyUI workflow is the node graph saved inside generated PNGs as prompt and workflow records. Learn the two JSON forms and how isGenAI reads them.
A1111 parameters
The AUTOMATIC1111 parameters record stores the prompt, negative prompt, steps, sampler, seed and model in a PNG text chunk or EXIF.
EU AI Act Article 50
Article 50 of the EU AI Act sets transparency duties: chatbot disclosure, machine-readable marking of AI output, deepfake labels.
Transparency obligation
A transparency obligation under the EU AI Act is a duty to inform people about AI use: interaction notices, output marking, deepfake labels.
Machine-readable marking
Machine-readable marking is the Article 50(2) duty to mark AI output so software can detect it: C2PA, IPTC metadata, watermarks.
Manifest
A C2PA manifest is the signed unit inside Content Credentials: assertions, a claim and a signature.
Signed receipt
A signed receipt is the Ed25519-signed record isGenAI issues for a check, binding the file hash, findings and time.
Hash
A hash is a fixed-length fingerprint of a file’s bytes; SHA-256 is the standard isGenAI uses.
Perceptual hash
A perceptual hash is a fingerprint of how an image looks, stable across resizing and compression.
Generator tag
A generator tag is a metadata field naming the software that made an image, such as XMP CreatorTool or EXIF Software.
Metadata stripping
Metadata stripping is the removal of EXIF, XMP, PNG chunks and C2PA from a file, by platforms, converters or privacy tools.
Screenshot laundering
Screenshot laundering is re-capturing an AI image to shed its provenance. Learn why it beats metadata checks, what survives and how to counter it.
Soft binding
A soft binding is a C2PA mechanism, such as an invisible watermark or a fingerprint, that reconnects content to its manifest after metadata is stripped.
Hard binding
A hard binding is the hash of the file content inside a signed C2PA claim; any change breaks it.
Trust list
A trust list is the set of certificate authorities a C2PA validator accepts, so a valid signature can still be untrusted.
Certificate
A signing certificate is the X.509 credential that identifies who signed a C2PA manifest. Learn about chains, trust lists and isGenAI checks.
How to use this term when checking a file
- Upload the original file to the AI image detector or the tool named on the page.
- Find the finding that uses this term under “How we checked” and read the source field it names.
- Weigh it by kind: a signed record, a declared field, a saved prompt and a matching copy are different strengths of evidence.
- Remember that absence of a record is not evidence of a human author.