FAQ

Frequently asked questions about AI image review

These answers explain what the product does, where its evidence comes from, and where human review still matters.

Can AI image detectors be wrong?

Yes. AI image detection is probabilistic, so false positives and false negatives can happen. That is why ImageVerity combines multiple evidence layers.

Can metadata prove an image is real?

No. Metadata can strengthen a review, but it can also be missing, stripped, or edited. It is useful evidence, not complete proof.

What does C2PA help with?

C2PA can provide signed provenance about image origin and editing history when trusted Content Credentials are present. It is a strong provenance signal, but it does not by itself prove that the depicted event is true or accurate.

Does Sightengine prove that an image is AI-generated?

No. Sightengine provides an AI-generation score based on image pixels rather than relying on EXIF, C2PA, or watermarks. That makes it useful when metadata is missing, but the score should still be treated as probabilistic evidence.

Why include Hive as a second detector layer?

Hive can add another classification and confidence-style signal for AI-generated or manipulated media. It is useful when teams want cross-checking before escalation, but disagreement between detectors should be preserved in the report.

What does VRE look for?

VRE looks for explainable visual cues such as repeated texture, inconsistent lighting, unusual geometry, broken text, unnatural hands, reflection mismatch, boundary artifacts, and scene-level inconsistency. These cues help explain risk; they are not standalone proof.

What if C2PA is missing?

Missing C2PA credentials are not a verdict. The file may have passed through tools or platforms that do not preserve provenance. In that case, reviewers should rely more on metadata, detector signals, VRE, and manual context.

Can screenshots or compression affect review?

Yes. Screenshots, recompression, resizing, and social-platform processing can remove metadata, break provenance, or change detector confidence. The report should note that limitation instead of overstating certainty.

Do you store uploaded images forever?

No. The product is designed around limited retention, and lower-tier flows clean up originals after a short period.

Is this suitable for legal proof?

No. ImageVerity should be used for risk judgment, prioritization, and operational review rather than as a single final legal determination.

Who is this product for?

The clearest fits are news verification, platform moderation, creator ecosystem review, and e-commerce or marketplace risk control teams.