About AuditImage

AuditImage is a free tool that reads a picture's metadata and judges where it came from. This page explains what it does, what it does not do, and why.

What it does

Upload a picture and AuditImage reads every metadata segment in the file: C2PA Content Credentials, EXIF, XMP, IPTC, PNG text chunks, and the implicit label required by the AI-generated content labelling rules. It then cross-checks them and gives a one-line verdict with itemised findings.

The verdicts: a signed AI-generation declaration, a signed camera capture, AI generation self-declared in metadata, contradictory metadata, a camera capture per EXIF, metadata with no origin claim, and no usable metadata.

What it does not do

  • No pixel-level AI image classification. Those tools guess with statistical models; that has nothing to do with metadata and their error rates are a separate matter.
  • No detection of SynthID or other invisible watermarks. They are not in the file's metadata.
  • C2PA timestamp tokens are reported as present, not verified. The signature, the content hash binding, the assertion hashes and the certificate chain are all verified, the chain against the official C2PA trust list and the CAI's interim list. For a second opinion use CAI Verify or c2patool.

Why metadata only

Metadata is the record the file wrote about itself, and the only thing that can be checked word for word. It can prove "the file says so", not "it is so"; but when parts of the file contradict each other, it can prove "someone tampered". AuditImage states that boundary clearly instead of producing a confident-looking percentage.

Privacy

Pictures are parsed in server memory, never written to disk, and discarded once parsed. Only the resulting report and the file's SHA-256 hash are stored. Report pages can be shared by link and are excluded from search engine indexing by default. See the privacy policy.

API

POST /api/inspect accepts a multipart upload and returns the same JSON report the page shows. There is an example at the bottom of the home page.

Technology

TanStack Start, React, SQLite. Metadata parsing uses ExifReader; the C2PA JUMBF and CBOR parsing and the segment-level JPEG, PNG and WebP walks are our own.