About the project
A provenance-first image review tool, not a black-box AI detector.
AI Image Evidence Checker helps people inspect evidence attached to an image file: C2PA Content Credentials, metadata, raw byte markers, camera-like signals, and frequency clues. It is designed to explain what is present, what is missing, and what remains inconclusive.
What it is for
- Creators checking whether a file still carries useful provenance before publishing.
- Journalists, moderators, and researchers triaging online images before deeper review.
- Teams explaining why an image report is inconclusive instead of forcing a fake-or-real label.
Evidence hierarchy
The checker treats cryptographically verifiable provenance as stronger than marker-only or content-derived signals. That hierarchy keeps the report useful without pretending every image can receive a definitive attribution.
- Layer 1
Trusted C2PA Content Credentials and asset binding when available.
- Layer 2
OpenAI-style or other provenance marker strings that need verification context.
- Layer 3
EXIF, XMP, camera-like formation clues, and byte-level marker evidence.
- Layer 4
Frequency-domain clues that can support review but are not standalone probabilities.
Limits and privacy posture
Missing metadata does not prove an image is AI-generated. Screenshots, reposts, edits, compression, and privacy tools can remove provenance. Frequency clues are forensic context, not legal attribution or a probability score.
Uploads are processed temporarily for the requested report and are not used for model training or persistent galleries in the current product flow.