Question 1

How does this tool know an image is AI-generated?

Accepted Answer

Its primary verdict only reports 'confirmed' or 'strong' when the file contains a real, checkable signature — a C2PA Content Credentials manifest, IPTC's official 'AI-generated' digital source type, a generator named in a structured metadata field, or a generator's own fingerprint (like Stable Diffusion's embedded prompt/seed data or a ComfyUI workflow).

Question 2

What is the Pixel & Noise Analysis section?

Accepted Answer

It's a separate, experimental, metadata-independent signal that runs even when a file has no AI metadata at all (common once a photo's been re-saved or shared). It converts the image to grayscale, extracts a noise-residual map with a high-pass filter, and re-measures noise uniformity at three different block scales (8px, 16px, 32px) so the reading isn't an artifact of one arbitrary scale choice. It also checks R/G/B channel noise correlation (real camera sensors leave correlated noise across channels via Bayer demosaicing), compares noise in busy/detailed regions against flat regions, and looks for an unusually dominant periodic frequency — all classic image-forensics techniques. It's a set of statistical hints, not a machine-learning verdict, and it's shown separately from the evidence-based result so the two are never confused.

Question 3

Why isn't the noise/frequency analysis as reliable as the metadata check?

Accepted Answer

Pixel-based AI detectors — including this one's noise/frequency signals — work by spotting statistical tendencies, not certainties. Plain backgrounds, heavy denoising, screenshots, and genuinely repeating real-world textures (fabric weave, brick, blinds) can all produce the same signals as AI generation, and newer generators keep closing these gaps. That's why it's labeled 'Experimental' and kept separate from the evidence-based verdict, which only speaks when it has something concrete to point to.

Question 4

What does 'inconclusive' mean?

Accepted Answer

It means no AI-generation signature was found in this file's metadata — not that the image is confirmed authentic. Social platforms, screenshots, messaging apps, and simple re-saves routinely strip all metadata, which erases this kind of evidence either way. Check the Pixel & Noise Analysis section for a secondary, lower-confidence signal in that case.

Question 5

What is C2PA / Content Credentials?

Accepted Answer

C2PA (Coalition for Content Provenance and Authenticity) is an open industry standard, backed by Adobe, Microsoft, Google, OpenAI, and others, for embedding tamper-evident 'Content Credentials' in media files — recording how an image was created or edited, including AI generation. This tool detects Content Credentials Manifest when present, but it doesn't cryptographically verify the manifest's signature, which requires the official C2PA trust-chain verifier.

Question 6

Is my image uploaded anywhere?

Accepted Answer

No. The entire file — including the pixel analysis, which runs on an in-memory canvas — is processed inside your browser. It never touches StatFlare's servers or any third party.

AI Image Detector

Frequently Asked Questions

More Free Tools