isGenAI.com

Glossary

Perceptual hash: matching copies that lost their metadata

A perceptual hash is a short fingerprint computed from an image’s visual content, such as the brightness pattern of a downscaled greyscale version, designed so that resized, recompressed or lightly edited copies of the same picture produce similar values that can be compared by Hamming distance.

Definitions and isGenAI behaviour as of October 2026. Updated .

See the term in a real reportFrequently asked questions
Code-drawn illustration for Perceptual hash

Illustration · PNG file

  1. Signature + IHDR
  2. iTXt XML:com.adobe.xmprecord
  3. tEXt parameters
  4. tEXt prompt / workflow
  5. tEXt invokeai_metadata
  6. eXIf
  7. caBX C2PA manifest
  8. IDAT pixels
  9. IEND

AI source type declared

The file labels itself as AI-made.

IPTC DigitalSourceType such as trainedAlgorithmicMedia is read from XMP and shown with its exact value.

Check an image →
AI source type declared: where this record sits inside a PNG file and what the isGenAI report says about it. The picture is a code-drawn illustration.

How it works

A difference hash shrinks the image to a tiny greyscale grid and records whether each pixel is brighter than its neighbour, producing a bit string. Two copies of the same picture differ in only a few bits even after a social platform re-encodes them; unrelated pictures differ in many. It is deliberately insensitive to the metadata that cryptographic hashes track, which makes it the tool for reconnecting a stripped copy to a known original.

Perceptual hashes can collide on very similar pictures and can be fooled by heavy crops, so a near match is reported as near, never as identical.

How isGenAI uses it

The checker computes an isgenai-dhash-v1 fingerprint for every image and compares it with its records and with soft-binding entries. An exact fingerprint match and a near match within a small Hamming distance are reported separately from an exact SHA-256 match, so the report always says which kind of copy was found. The fingerprint algorithm and dimensions are printed in the evidence so the result can be reproduced.

Why it matters for screenshots

A screenshot of an AI image keeps its look while losing every record. If the original was checked before and carried a credential or a saved prompt, the fingerprint can link the screenshot to that record. This is the only route from a stripped copy back to file evidence.

How to use this term when checking a file

  1. Upload the original file to the AI image detector or the tool named on the page.
  2. Find the finding that uses this term under “How we checked” and read the source field it names.
  3. Weigh it by kind: a signed record, a declared field, a saved prompt and a matching copy are different strengths of evidence.
  4. Remember that absence of a record is not evidence of a human author.

See the term in a real report

Frequently asked questions

Is a perceptual hash the same as SHA-256?

No. SHA-256 changes on any byte; a perceptual hash stays similar across resizing and compression. isGenAI uses both and reports which matched.

Can a perceptual match prove AI origin?

Only indirectly: it links your copy to a record that may carry a credential or saved prompt. The report shows that record’s evidence.

How close is a near match?

Within a small Hamming distance on the fingerprint bits; the report prints the relationship as near rather than exact.

Does cropping defeat it?

Heavy cropping can. Light edits, resizing and recompression usually do not.

Sources

Definitions follow the published standards linked above; the isGenAI behaviour described is the deployed checker as of October 2026.