Plain answers about what we analyse, what a finding means, why a result can be inconclusive, how calibration and provenance work, how records are sealed, and what we do not claim.
Illustration only. No real customer data or recorded demo is shown.
Still images in common formats, including JPEG, PNG, WebP, and HEIC. We look for signs of synthetic generation, supported face swaps, and supported image manipulation.
Voice and video files are accepted for analysis, but they are in testing and return inconclusive. They do not produce production decisions.
A finding names what was detected: AI-generated, face-swap, or manipulated. It is released only when the evidence meets the calibrated threshold for your use case.
A finding is about the image. It does not say who a person is, what they intended, or whether they did anything wrong.
The evidence did not meet the operating threshold, the use case has not been calibrated yet, or the media type is still in testing. In each case no finding is released.
Inconclusive is not a statement that the image is authentic. The case should go to human review.
Each use case has its own operating point, because a hiring review and a media screen carry different consequences. A use case moves from uncalibrated, through calibration, to calibrated, and results ship only once it is calibrated.
Your team's review of a case can be recorded as a label that feeds future calibration. A label never changes the sealed result.
Where an image carries Content Credentials, we check the C2PA signature and trust chain and show whether they verified, failed, or were absent.
Provenance is context alongside the forensic analysis. It never decides a result on its own.
When an analysis completes, the case, finding, evidence, C2PA status, and time of decision are sealed into a record with a SHA-256 hash. Records are append-only.
Export a record through the API and re-check its hash at any time: if the record has changed, the hash will not match.
We do not confirm identity. We do not publish accuracy percentages yet. We make no claim about legal admissibility. We do not return voice or video verdicts, and we do not protect live calls or meetings today.
We show a certification or compliance badge only when current documentation exists and can be supplied to a buyer.
Bring your own known-real and known-manipulated images to a demo and see where the system finds, abstains, and why.
