Prepare the asset
The file is validated, uploaded, and prepared for detector, metadata, and provenance analysis.
How it works
How ImageVerity reviews an image
ImageVerity is built to combine multiple evidence layers into one readable workflow, so reviewers can understand why an image looks suspicious instead of trusting one opaque score.
Read metadata
ExifTool reviews EXIF, XMP, IPTC, ICC, maker notes, and related fields for capture details, edits, software traces, and unusual metadata patterns.
Check provenance
c2patool verifies C2PA Content Credentials and signed issuer chains when they exist, then records whether provenance is valid, missing, or broken.
Run Sightengine and Hive signals
Sightengine contributes a pixel-level AI-generation score, while Hive can add a secondary classification and confidence-style cross-check for escalation decisions.
Add VRE visual reasoning
VRE looks for explainable visual cues such as texture repetition, lighting mismatch, geometric inconsistency, strange hands, broken text, reflection errors, and object boundary artifacts.
Summarize carefully
The report turns those layers into a risk summary, separates aligned evidence from conflicting evidence, and recommends a next step for the reviewer.
Escalate when needed
High-risk or conflicting outputs should move into manual review instead of being treated as absolute proof.
Result framing
Built for operational review, not absolute truth claims
A strong signal can justify escalation, moderation, or deeper investigation, but it should not be described as infallible proof that an image is real or fake.
That is especially important when provenance is missing, metadata is stripped, or detector layers disagree.