Check for AI in four steps.
isGenAI reads the evidence: saved creation records, explicit AI labels and signed credentials, checked against the file itself. Start with the original content, then read the findings with their sources. Free, no account, for every public check.
1. Start with the source
Upload the original image or paste a public link, text or code. The original download carries the richest records, so keep it separate from any screenshot or repost: the copy you choose decides which saved records the check can read.
2. Read every available record
The checker reads creation records, AI labels and signed credentials, and matches the file against our records. Text and code give up explicit statements that AI was used. When an image-scan estimate is available, it is labelled as an estimate, apart from a recorded declaration.
3. See what finished
Open ‘How we checked’ to read every finding and every completed check. A check that did not run is listed as exactly that, apart from a completed check with a clean result. A saved prompt, software name or label is shown as the recorded, editable claim it is.
4. Decide on evidence
‘No AI evidence found’ means the completed checks read this copy and found no AI record: a verified clean copy, reported as exactly that. Conflicting records are flagged for your review. Your feedback steers what we investigate next; it never silently changes a result or becomes a training label.
Image input and privacy
The main image check accepts PNG, JPEG, WebP, AVIF and single-image TIFF files up to 20 MiB (20,971,520 bytes). TIFF uploads are classic single-page files with no subimages or pyramids and at most 40 million pixels; multipage TIFF and BigTIFF are rejected up front rather than silently converted.
Main-checker uploads are processed on the server and private by default. Public URL checks create public, signed receipts by default. When enabled, external detection or prompt-reconstruction services receive the submitted content, and the privacy details spell out exactly when. Read them before checking sensitive material.
Follow the image-checking guide for supported files, result examples and keeping a report.
Pick the right tool for the question
- Read camera, date or GPS fields: the EXIF viewer reads metadata locally in your browser, so the file never leaves your device. A saved field is a recorded claim, and the viewer shows it as one.
- Read saved generation instructions: the image prompt extractor returns the exact saved prompt, and the ComfyUI workflow viewer returns the full graph JSON. Both read what the file holds; a reconstructed prompt is always labelled as a new approximation.
- Verify a signed record: the C2PA checker validates files up to 25 MiB in memory, without storing them, and reports signature validity and signer trust separately, the way the standard intends.
- Compare an original and an export: metadata comparison shows field by field which saved prompt, model and workflow records survived. EXIF, GPS and C2PA signatures have their own dedicated tools above.
- Resize or convert an image: the browser image tools edit pixels on your device. Re-read the output’s metadata afterwards if preserving records matters.
Try a check with a known result
Explore the controlled image metadata examples for downloadable files, exact observations and reproduction code. The signed and unsigned C2PA controls have identical decoded pixels and different signature records, so you can see exactly what a signature changes. These examples demonstrate reader behavior, reproducibly, with published hashes.
Read the image history guide to interpret recorded claims, or report a reproducible problem without sharing private content.