AI Image Watermark Detector
Drop an image and DetectMark parses it in this tab: Content Credentials, EXIF, GPS, XMP, PNG text chunks and generator tags, each with its own verdict.
- Watermark type
- File metadata
- Hidden characters
- No
- Official detector
- Available: Available Parsed locally in the browser
- DetectMark support
- DetectMark support: Available Reads and strips locally
- Removal
- Available: Available Removed and verified by rescan
- Watermark type
- Pixel watermark + C2PA metadata
- Hidden characters
- No
- Official detector
- Available: Available Verifiable via OpenAI’s Verify tool and API
- DetectMark support
- DetectMark support: Available Reads C2PA locally; OpenAI check on consent
- Removal
- Partial: Partial Metadata only; pixel mark stays
What a file can tell you
Most of what identifies an image as generated is written down in the file, not hidden in it. A C2PA manifest names the generator and carries a signature. XMP can carry a DigitalSourceType assertion. PNG text chunks are where Stable Diffusion stores the prompt. EXIF holds the camera, the timestamps and sometimes the coordinates.
DetectMark reads all of those and prints the values as text, never as markup, so a crafted file cannot do anything but be read.
What a file cannot tell you
An image with no metadata is not evidence of anything. Social platforms strip metadata on upload, screenshots have none to begin with, and an export from almost any editor drops most of it. That case reports Unknown, which is the accurate answer, rather than Clean.
The reverse trap matters too. A C2PA manifest is not proof that a picture is AI-generated. The format is equally capable of asserting that a camera captured it. The tool shows the assertions it found and lets them speak.
Pixel watermarks
SynthID and its relatives live in the pixel values, and verifying them needs the provider's own detector. In the local scan every pixel-watermark row reports Unknown: this tool never guesses from the file's look, and it never infers a verdict from the presence of C2PA. For OpenAI-generated images, this page can send the original file to OpenAI's Content Provenance API after one explicit consent click, and it shows OpenAI's SynthID and Content Credentials results as they come back. That covers OpenAI-supported signals only, and nothing is sent until you confirm. A not-detected result says nothing about other providers or about a file whose marks were stripped. Google's own SynthID Detector portal remains waitlist-only.
FAQ
C2PA manifests, XMP including DigitalSourceType, EXIF and GPS, camera and software tags, and PNG text chunks including Stable Diffusion prompts.
Not in the browser. This tool reports the C2PA that usually accompanies it, and marks the pixel row Unknown. For Google images, pixel verification runs through Google Search surfaces and Chrome. For OpenAI images, one consent click here sends the original file to OpenAI's Content Provenance API, which verifies the SynthID and C2PA OpenAI embeds, for OpenAI-originated marks only.
Not for the scan or the cleanup: all parsing runs locally on the File object, and zero bytes of the image are uploaded. The optional OpenAI check is the one exception. It sends the original file to api.openai.com only after you confirm in a dialog that states the exact byte count, and the counter then shows those bytes.
The verdict is Unknown for the metadata check. That does not mean the image is or is not AI-generated; it means the file carries no readable label.
No. C2PA can also assert human capture. The tool reads the assertions and shows them; a "Made with AI" assertion is one of many possible claims.
Every check runs in your browser: how it works. The limits are written down too: what DetectMark cannot remove.