02Method
How Fileverity reaches a conclusion
No single test can prove an image is authentic. Fileverity collects independent lines of evidence, weighs them against each other and shows its work.
01
Five stages of every analysis
- 01 / 05
File intake
The file is hashed with SHA-256 on arrival, so every finding is tied to the exact bytes received. Format, dimensions, orientation and compression history are recorded.
- 02 / 05
Provenance and metadata
Content Credentials (C2PA), EXIF and XMP metadata, generation parameters written by AI tools, machine-readable AI labels and known invisible watermarks are extracted and validated.
- 03 / 05
AI-generation analysis
Independent detectors examine the pixels for statistical traces left by generative models. Their outputs are calibrated and combined, so no single detector can dominate the result.
- 04 / 05
Manipulation analysis
Independent forensic analyses look for areas that were pasted in, cloned, retouched or regenerated, and mark them on the image. They also examine compression history and pixel consistency, and read the editing-software history recorded in the file.
- 05 / 05
Evidence fusion and report
All findings are combined into a conclusion, an authenticity score and a risk level. Conflicts between sources are flagged, and the full evidence table is written to the report.
02
Principles behind the method
- 01 / 04
Evidence over verdicts
Every conclusion lists the findings that support it and where they came from, so a reviewer can follow the reasoning.
- 02 / 04
Hard evidence first
Valid Content Credentials and recognized watermarks count as strong evidence. Pixel analysis adds weight but never silently overrides a cryptographic claim.
- 03 / 04
Honest uncertainty
When signals disagree or are too weak, the answer is inconclusive. A confident-looking score is never forced.
- 04 / 04
Edits are not always fraud
Saving a photo in an editor is not the same as changing what it shows. Fileverity reports software history and content changes separately.
03
Risk levels and recommended actions
Results map to six risk levels, from L0 to L5. Each level carries a recommended action, so a team can route every image the same way. A stricter policy is available for high-stakes workflows.
- 01L0: request the original file and check again
- 02L1 – L2: accept
- 03L3 – L4: review manually before deciding
- 04L5: do not accept
04
Frequently asked questions
Q01What is the authenticity score?
The authenticity score is a number from 0 to 100 that expresses how strongly the evidence supports the image as an authentic, unedited capture. It is shown in steps of 5 and is lower when the file limits what can be established, for example after heavy recompression.
Q02What are Content Credentials?
Content Credentials are signed provenance records, defined by the C2PA standard, that can state how an image was captured, edited or generated. Fileverity validates the signature and compares the claims with the image itself.
Q03Why can a result be inconclusive?
A result is inconclusive when the evidence is too weak or contradictory to support a conclusion, and Fileverity says so instead of guessing. The report explains what is missing, such as an original file or intact metadata.
Q04How do you verify a Fileverity report?
Scan the QR code on the report or open its verification address. The verification page confirms that Fileverity issued the report and shows its key results and fingerprints. It can also check, on your own device, whether an image file matches the report.
Put evidence behind every image decision.
Tell us about your workflow and volume. We will set up access and walk your team through real cases.