Skip to content
Delete Image Metadata

AI Metadata Checker

See exactly which AI markers an image carries before you publish it. This tool only reads — nothing is changed, and nothing is uploaded.

  • .jpg
  • .jpeg
  • .png
  • .webp
Free
No account, no limit on files
0
Files uploaded — processing is local
4 formats
Lossless — pixels never re-encoded
Processed in your browser

How it works

  1. 1

    Load an image

    JPG, PNG or WebP. The file is read in your browser.

  2. 2

    Read the report

    Each marker is listed with where it lives in the file and what it says.

  3. 3

    Decide what to do

    Publish as is, or send the file to the remover to strip the markers.

What this checker looks for

The check is a byte-level read of the file's metadata blocks. Each marker below is something a generator or editor physically wrote into the file, so a positive result is evidence rather than inference.

  • PNG text chunks named 'parameters', 'sd-metadata' or 'dream' — the prompt, negative prompt, steps, sampler, CFG scale, seed, size and model hash
  • PNG chunks containing a ComfyUI node graph as JSON, often the largest block in the file
  • C2PA content credentials in a JPEG APP11 JUMBF box or a PNG caBX chunk
  • An IPTC digital source type in XMP declaring the content as algorithmically generated
  • Generator names recorded in ordinary Software or CreatorTool tags — Firefly, Midjourney, DALL·E, InvokeAI, NovelAI, FLUX and others

What a clean result does and does not mean

If the checker reports no markers, that means the file's metadata contains no evidence of AI generation. It does not mean the image was not generated. An AI image whose metadata has already been stripped — by a metadata cleaner, a social platform's re-encode, or a screenshot — looks identical to a camera photo at this level.

This tool reads bytes. It does not analyse the picture, and it deliberately does not pretend to: an honest 'no markers found' is more useful than a confident guess.

Why check before publishing

The practical reason is that prompts are often more revealing than people expect. They contain client names, project codenames, unreleased concepts, and the specific phrasing a creator has spent time developing — all of it published to anyone who opens the file.

There is a second reason for anyone receiving images rather than making them: checking a file tells you what it claims about its own origin. That claim is editable, so it is a signal rather than proof, but its absence and its presence are both worth knowing.

What this tool does

  • Parses Stable Diffusion parameters into named fields rather than dumping raw text
  • Detects C2PA content credentials in both JPEG and PNG containers
  • Read-only: your file is never modified by this page

What it does not do

  • Cannot tell whether an image is AI-generated when the metadata has been stripped
  • Does not analyse the picture itself and does not run any classifier
  • Cannot detect pixel-encoded watermarks such as SynthID

AI Metadata Checker

See exactly which AI markers an image carries before you publish it. This tool only reads — nothing is changed, and nothing is uploaded.

Open the tool

Reviewed and updated . Behaviour described on this page is covered by the project’s automated tests.

Frequently asked questions

Can this tell me if an image was made by AI?

Only when the file still carries a marker saying so. If the metadata has been removed, no metadata-based check — including this one — can tell the difference between a generated image and a photograph.

Where does Stable Diffusion store the prompt?

In a PNG text chunk, usually keyed 'parameters'. It is plain text: the prompt, negative prompt, sampler, steps, CFG scale, seed and model hash all sit there readable by anyone who opens the file.

Does checking upload my image?

No. The file is read with the browser's File API and parsed in page memory. You can confirm it in your browser's network tab.

What is a JUMBF box?

The container format C2PA uses to store content credentials inside an image. In a JPEG it lives in an APP11 segment; in a PNG it sits in a caBX chunk.

Why does my image show a generator name but no prompt?

Many tools write only a Software or CreatorTool tag. Prompt storage is mostly a Stable Diffusion interface behaviour — hosted services such as Midjourney and DALL·E typically do not embed the prompt in the file you download.

Does the checker modify my file?

No. This page only reads. Removal is a separate tool, and your file is never written to.

Can I check an image someone else made?

Yes, and that is a common use. Bear in mind metadata is editable, so treat a result as a claim rather than proof.

What if the check finds nothing?

That means the file carries no AI markers. It does not mean the image was not generated — stripped or re-encoded files look identical at this level.

Does it check for SynthID?

No. SynthID is encoded into pixel values rather than metadata, and reading it requires Google's own detector.

Keep going

Related tools

Back to the main metadata remover