How this AI image evidence checker works
Check image provenance without pretending one signal is perfect.
Understand how the checker ranks provenance evidence, what supporting signals can still tell you, and why missing metadata should usually be read as inconclusive rather than proof.
Evidence first, not a simplistic verdict
This free AI image checker reviews file evidence rather than returning a simplistic real-or-fake label. It checks C2PA Content Credentials, OpenAI-style provenance markers, EXIF and XMP metadata, raw byte markers, camera-like formation clues, and frequency-domain signals.
The goal is to help you understand what evidence is present, what is missing, and what should be treated as inconclusive. That framing is safer than pretending one detector score can settle attribution on its own.
What counts as the strongest result
The strongest result is trusted provenance: a valid C2PA manifest, trusted signature, and asset binding that matches the analyzed file. That is materially stronger than marker-only findings or content-based heuristics.
Marker-only findings, camera-like metadata, and frequency scores can still be useful, but they are supporting signals. Missing C2PA data does not mean an image is fake, because screenshots, social reposts, privacy tools, and older workflows often strip metadata.
How to interpret weaker signals
OpenAI-style markers and raw byte matches can explain why a file deserves more scrutiny, but they are not the same as cryptographic verification. EXIF and camera-like clues may support a camera-origin interpretation, yet they can also be absent, edited, or removed.
Frequency-domain outputs should be read as uncalibrated forensic context. Compression, edits, resizing, screenshots, and model changes can all shift those scores, so they should never be treated as a standalone probability of AI origin.
Next step
If you want signal-by-signal explainers before reading a report, use the guide library for C2PA, Content Credentials, OpenAI-style markers, EXIF metadata, screenshots, byte markers, and frequency-analysis limitations.