AI Image Metadata Checker
Open an image to read the metadata that can reveal how it was made.
Runs in your browser. Nothing is uploaded.
Reading the file…
This tool reads what the file says about itself. It is not an AI detector: it does not look at the pixels or invisible watermarks, and an image with no metadata can still be AI-generated.
This file contains a GPS location. Use before you share it.
| Format | |
|---|---|
| Size | |
| File | |
| Blocks |
Content Credentials (C2PA)
A C2PA manifest was found. This page reads its labels but does not check the signature, so it cannot tell whether the manifest is genuine or was changed. Use a C2PA verifier for that.
| Stored in | |
|---|---|
| Size | |
| Claim generator | |
| Software agent | |
| Actions | |
| Source type | |
| Signature box |
| Key | Value |
|---|---|
| Group | Field | Value |
|---|---|---|
Raw XMP packet
The EXIF reader could not load, so EXIF and IPTC fields are missing. Reload the page to try again.
Not a detector: invisible watermarks such as Google SynthID live in the pixels, not in metadata, and are not checked here.
The image is redrawn on a canvas and saved again, which drops EXIF (including GPS), IPTC, XMP, PNG text chunks and C2PA manifests. Colours are converted to sRGB, rotation from EXIF is applied, and animations keep only the first frame. JPEG and WebP copies are re-compressed at quality 92.
We re-read the new file: it has no EXIF, XMP, IPTC, text chunks or C2PA data.
Removing metadata does not remove watermarks drawn into the pixels. Some sites add their own metadata again when you upload.
What this tool reads
Open an image and the checker lists everything in its metadata that says how it was made:
- Generation settings in PNG files. AUTOMATIC1111, Forge and other Stable Diffusion web UIs write the prompt, negative prompt, seed, sampler and model into a
parameterstext chunk. ComfyUI stores the whole node graph inpromptandworkflow, NovelAI usesDescriptionandComment, and InvokeAI, SwarmUI and Fooocus have their own keys. The tool reads all three kinds of PNG text chunk, including compressed ones. - The same data in JPEG and WebP. Stable Diffusion web UIs put the parameters into the EXIF UserComment field when they save JPEG or WebP.
- IPTC Digital Source Type. Google, Meta, Midjourney and stock sites mark generated images with the value
trainedAlgorithmicMedia, and AI-edited ones withcompositeWithTrainedAlgorithmicMedia. - EXIF Software and XMP CreatorTool, compared with a list of known AI image tools.
- Content Credentials (C2PA). OpenAI, Adobe Firefly, Microsoft and some cameras attach a signed C2PA manifest. The tool finds it in JPEG APP11 segments, PNG
caBXchunks and WebPC2PAchunks and shows the claim generator, actions and source type it declares.
It also shows the camera make and model, warns about GPS coordinates, and lists every EXIF, IPTC and XMP field it found.
What it cannot tell you
- Missing metadata proves nothing. Instagram, Facebook, X, WhatsApp, Telegram and most other platforms strip metadata from uploads, and so do screenshots. Many generators write none at all. An AI image downloaded from social media will usually show "No metadata".
- Metadata can be faked. Anyone can add a camera model to a generated image or a prompt to a real photo with free tools. Treat every field as a claim.
- C2PA signatures are not verified. The tool detects a manifest and reads its labels, but it does not check the cryptographic signature or the certificate chain. To confirm a manifest is genuine and unchanged, open the file in a C2PA verifier such as the Content Authenticity Initiative's Verify site.
- Invisible watermarks are not checked. Google SynthID and similar marks live in the pixels and can only be read by their owners' tools.
Remove metadata before you share
Switch to Remove metadata to save a clean copy. The image is redrawn on a canvas and encoded again, which drops EXIF (including GPS location), IPTC, XMP, PNG text chunks and C2PA manifests. The tool then reads the new file back and confirms that nothing is left.
Re-encoding has side effects: colours are converted to sRGB, JPEG and WebP copies are compressed again at quality 92, and animated images keep only their first frame. Choose PNG if you need a lossless copy. Removing metadata also removes Content Credentials, which some publishers and platforms use to show where an image came from.
For resizing and converting many images at once, use the Image Resizer, which also strips metadata. To write better prompts for an image model, try the AI Image Prompt Builder. Tools for checking and protecting content are listed under Security and Images.
Can this tell me if an image is AI-generated?
Not with certainty. It shows what the metadata says: generator settings, AI source types, software names and Content Credentials. If an AI image has no metadata, which is common after it has been posted online, the tool has nothing to report.
How do I see the prompt of a Stable Diffusion image?
Open the original PNG, not a copy downloaded from a social network. If it was saved by AUTOMATIC1111, Forge, ComfyUI or a similar tool, the prompt, negative prompt and settings appear in the Generation data panel with a button to copy them.
What is trainedAlgorithmicMedia?
It is a value from the IPTC Digital Source Type list. It means the creator declared that the image was made with a generative AI model trained on sample content. Platforms such as Google and Meta use it to label AI images.
Is my image uploaded?
No. The file is read in your browser. Nothing is sent to our server, so you can check private photos safely.
Does removing metadata make an image undetectable?
No. It removes the labels in the file, not watermarks in the pixels, and detection tools that analyse the image itself are unaffected. Use it to protect your privacy, for example to drop GPS coordinates from a photo, not to hide where an image came from.
Security tools from our directory
Create stunning AI-generated images from text descriptions.
AI-driven image and video editing with natural language commands.
