AI IMAGE
DETECTOR

Evidence deskLocal / private / image-only

Inspect the
evidence.
Not the vibe.

An evidence-based AI image detector for Content Credentials, generator metadata, and workflow traces. Your image stays in this tab.

Read the limits
LOCAL EVIDENCE READER
ENGINE 0.2

INPUT / 01

Drop one image
on the evidence desk.

PNG, JPEG, WebP, GIF, AVIF or HEIC · up to 100 MB

Choose image

METHOD / NOT MAGIC

One question.
Four different mechanisms.

  1. 01

    Validate the signed record

    The official open-source C2PA engine checks manifest structure, signature, asset binding, and configured trust anchors.

  2. 02

    Read what exports leave behind

    PNG text, EXIF, XMP, IPTC, workflow JSON, generator names, prompts, and software fields are inspected locally.

  3. 03

    Name the blind spots

    A missing credential cannot exclude SynthID, a removed metadata block, a screenshot, or a model-based visual signal.

READ BEFORE USING THE RESULT

“Nothing found” means no supported evidence was present. It does not mean “made by a human.”

Provenance is evidence about a file and its history, not a truth score for what the pixels depict. Editing, screenshots, re-exporting, and platform processing can change or remove inspectable evidence.

Open the field guide

FIELD NOTES / FAQ

Clear answers, without the theatre.

Does the image leave my device?

No. The selected bytes are read by JavaScript and the C2PA WebAssembly module in this browser tab. The WASM engine and the release's public C2PA trust-list snapshot are served as site assets; your image is not sent with those requests.

Can this prove an image is AI-generated?

A trusted credential or explicit generator metadata can provide strong evidence about the file. An empty result cannot prove human authorship because evidence can be absent, stripped, or stored in a proprietary watermark.

Does this detect Google SynthID?

It reports inspectable Google or SynthID references but cannot reproduce Google’s closed pixel-level detector. Use Google’s supported verification path when SynthID is the exact question.

Why show invalid credentials?

An invalid or tampered credential is important evidence. It means a manifest was found but did not pass the verifier’s structural, cryptographic, or asset-binding checks.