Choose an image, click Check for AI Marks, then read the verdict and the evidence table under it.
Choose an image, then click Check for AI Marks. The scan runs in your browser - the file is never uploaded.
Example result - what a DALL-E 3 download looks like in this checker:
| Mark | Where | Value |
|---|---|---|
| Content Credentials (C2PA) | JUMBF data in the file | c2pa manifest present |
| IPTC digital source type | XMP metadata | trainedAlgorithmicMedia (AI-generated) |
| Generator signature | C2PA claim | OpenAI / DALL-E |
AI Watermark Detector
Check whether an image carries the provenance marks AI generators write into their files. The checker reads the file's own bytes in your browser and lists every mark it finds: Content Credentials (C2PA), the IPTC "AI-generated" label, generator signatures such as DALL-E or Adobe Firefly, and Stable Diffusion prompt chunks.
Downloads from DALL-E 3, GPT-4o, and Sora carry a C2PA manifest plus the IPTC "trainedAlgorithmicMedia" label. Adobe Firefly names itself in the metadata. Stable Diffusion and ComfyUI embed the generation prompt inside PNG text chunks. Each found mark is shown as an evidence row - what it is, where it sits, what it says.
A clean result needs more care than a found one. These marks live in metadata, so a screenshot or re-save removes them, and Gemini's SynthID watermark hides in the pixels where only Google's own checker can read it. "No marks found" means exactly that - never proof that a person made the image.
The scan is a byte-level read of the file you choose: nothing is uploaded and there is no model download. To see every tag an image carries - camera, date, GPS - use the image EXIF metadata viewer. Checking writing instead of an image? The AI text detector scans text for machine-writing signals.
Frequently Asked Questions
Can this tool tell for sure whether an image is AI-generated?
No tool can. This checker reads the provenance marks generators write into the file, so a found mark is solid evidence of AI origin. But those marks are removable - a screenshot or re-save strips them - and Google's SynthID watermark is invisible to any metadata scan. A clean result means "no readable marks survived", never "made by a human".
What are Content Credentials (C2PA)?
An open standard for recording who or what created a file. OpenAI, Adobe, and camera makers write a signed C2PA manifest into the image; this page detects the manifest container and any generator names inside it. Full cryptographic verification of the signature chain needs the official Content Credentials Verify site, which the result links to.
Can it detect Gemini or Imagen images watermarked with SynthID?
No. SynthID is hidden in the pixels themselves rather than the metadata, and only Google's own checker can read it. If a Gemini image still carries its metadata, this page can catch the generator name there, but the SynthID mark itself is out of reach for any browser tool.
Is my image uploaded anywhere?
No. The file is read with the browser's own FileReader and scanned locally; the page works the same if you go offline after it loads. Nothing about the image or its result leaves your device.
Which formats does it understand?
JPEG, PNG, and WebP are scanned structurally - C2PA containers, XMP packets, and PNG text chunks are located by walking the file format itself. TIFF and HEIC fall back to an EXIF read plus a byte-pattern check, which still catches most marks but is less thorough.
Can I remove an AI watermark from an image?
Metadata marks disappear on any re-save, but removing them defeats labels that exist so viewers know an image's origin - AI-content labeling is legally required in a growing number of places, and stripping ownership marks from someone else's image can violate copyright law. This site does not offer an AI-provenance remover for that reason.