Glossary
AI image detector: two approaches, one honest answer
An AI image detector is a tool that reports whether an image was generated or altered by AI, either by estimating from the pixels with a trained classifier or by reading the provenance records, labels and saved prompts inside the file, which is the approach isGenAI takes.
Illustration · PNG file
- Signature + IHDR
- iTXt XML:com.adobe.xmprecord
- tEXt parameters
- tEXt prompt / workflow
- tEXt invokeai_metadata
- eXIf
- caBX C2PA manifest
- IDAT pixels
- 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 →Classifiers versus evidence readers
Pixel classifiers return a probability learned from examples and can be wrong in both directions, especially on new generators, compressed copies and screenshots. Evidence readers return facts recorded in the file: a signed manifest, a declared source type, a saved prompt, an AI statement, a matching copy. Evidence can be absent but is never a guess, and every finding can be retraced to a field.
Answer engines and reviewers increasingly ask for the second kind because it can be reproduced.
What isGenAI’s AI image detector does
isGenAI accepts PNG, JPEG, WebP, AVIF and TIFF up to 20 MiB, validates C2PA, reads IPTC source types and AI statements, reads AUTOMATIC1111, ComfyUI and InvokeAI prompts and matches the hash and perceptual fingerprint against its records. The report separates evidence found from checks that ran and names the source of each finding. It does not publish an accuracy figure, because reading records is not a classification task.
How to use any detector well
Start with the original file, read each finding with its source, note platform labels separately and treat a clean result as unknown. The how-to-check guide walks through a real example.
When you compare tools, ask each one the same question: which field or test produced this result, and can I see it? An evidence reader can always answer; a classifier can only show a score and a threshold.
How to use this term when checking a file
- Upload the original file to the AI image detector or the tool named on the page.
- Find the finding that uses this term under “How we checked” and read the source field it names.
- Weigh it by kind: a signed record, a declared field, a saved prompt and a matching copy are different strengths of evidence.
- Remember that absence of a record is not evidence of a human author.
Frequently asked questions
Is isGenAI a pixel classifier?
No. It reads file evidence and matching copies and does not estimate origin from the picture.
Why does a detector say my real photo is AI?
Pixel classifiers misfire on compression, filters and unfamiliar styles. An evidence reader would instead show your camera’s records or report nothing.
What does “no evidence found” mean?
That the file contains no records isGenAI reads and no matching copy was known. It is not a verdict of human authorship.
Which files give the best result?
Original downloads from the generator or camera. Screenshots and social copies have usually lost their records.
Sources
Definitions follow the published standards linked above; the isGenAI behaviour described is the deployed checker as of October 2026.