Glossary: image and document forensics
The terms that recur in reports, articles and conversations about AI fraud and forensic analysis, explained briefly and plainly.
Updated 2026-08-31 · Veriflied
Images
AI-generated image
An image made from scratch by an image model (GPT-image, Nano Banana, Midjourney, Stable Diffusion). No camera was involved.
AI editing / inpainting
A real photo where a region was changed or added by an AI model. The rest of the image is genuine. Harder to detect than fully generated images.
Deepfake
Video or image where a face or a voice was replaced or synthesised. Used broadly, but technically a subset of AI manipulation.
EXIF
The standard for metadata in photos: camera, time, settings, GPS. Read by programs, not visible in the image.
MakerNotes
Manufacturer-specific metadata fields (Apple, Samsung, Canon). Hard to fabricate, and their presence or absence says a lot.
Heatmap
An overlay showing which regions of an image the analysis assesses as manipulated.
C2PA / Content Credentials
A standard for cryptographically signed provenance. Present in some cameras and programs. Absence proves nothing; presence is strong.
Recompression
When an image is saved again as JPEG. Happens when sharing through apps. Leaves traces the analysis distinguishes from actual manipulation.
Documents
Producer / Creator
PDF metadata about which program made the file, and which program the document was written in.
XMP
A metadata packet in PDFs and images with, among other things, editing history (created, saved, edited) and timestamps.
Revision (incremental save)
When a PDF is saved again without being rewritten. The earlier version still sits in the file and can often be recovered.
Document id
An id pair in PDFs: one from creation, one from the latest save. Different ids mean the file was saved again.
TouchUp
A marker Adobe Acrobat leaves when text was edited directly in the PDF.
Issuer identity
The traces a real billing system or portal leaves: its own producer name, its own fonts, a fixed structure. If missing, the document was not issued by the system it claims.
Template generator
Websites that make fake receipts and invoices from ready-made designs. The result is a browser or HTML-to-PDF file without issuer identity.
Analysis and report
Forensic analysis
Systematic examination of a file's content, metadata and structure to determine whether it is what it claims to be.
Combination pattern
Several weak traces that together form a known fraud signature, for example 'freshly generated document without issuer identity'.
False positive
A genuine image or document wrongly flagged. Thresholds are calibrated so it happens rarely, and every finding is explained so it can be dismissed quickly.
Chain of custody
Documentation that the analysed file is identical to the submitted one, typically via a hash value (SHA-256).
Portfolio scan
Automatic analysis of many historical cases via API to find what nobody caught the first time.
Frequently asked questions
What is the difference between AI-generated and AI-edited?
Generated means made from scratch by a model. Edited means a real photo was partially changed. Editing is harder to detect and more common in claims.
Is missing metadata a sign of fraud?
No. Sharing through apps strips metadata. What counts is the pattern of what is missing, and the traces left in the image itself.
What does a risk score mean?
An overall assessment of the findings, not proof. The report always shows which findings drive the score and what to ask about.
Try it on your own document
Create an account and analyse your first 10 images or documents for free. You get a report with the concrete findings, not just a score. All processing in the EU.