AI Image Checker
Check whether an image carries AI generation data — by reading what is actually stored in the file, not by guessing.
Before you use this: Be sceptical of any tool that returns a confidence percentage for an image with no metadata. There is no browser-side model that can produce one, so the number is invented. This tool tells you what the file contains and lets you draw the conclusion.
Reading the file…
Findings
File
How to use this tool
- Select your file, drag it onto the box above.
- Processing starts automatically — there is nothing to configure.
- Check the result panel, which shows the original size, the final size and the page count.
- Click Download to save it.
This reports findings, not a verdict with a percentage. Where the file contains generation parameters, a provenance manifest, or an IPTC AI tag, that is direct evidence and it is shown to you in full. Where the file has been stripped — which is what every social platform does on upload — the honest answer is "inconclusive", and that is what you get.
The specification
| PNG text chunks | Stable Diffusion, Automatic1111 and ComfyUI write the full prompt, model, seed and sampler into the file |
|---|---|
| C2PA / Content Credentials | Signed provenance manifests from OpenAI, Adobe Firefly, Google and some cameras |
| XMP digitalSourceType | The IPTC standard tag "trainedAlgorithmicMedia" |
| EXIF | Camera make, model, lens, exposure, GPS |
| Formats | JPG, PNG, WebP |
| What it cannot do | Analyse pixels. It reads metadata only. |
Why uploads get rejected
Meeting the size is only part of it. These are the failures that actually come up:
- The image came from Instagram, WhatsApp, X or a screenshot. All of these strip metadata, so there is nothing left to read — for any tool, not just this one.
- The image was re-saved in an editor, which usually discards the original generation data.
- The file is a JPEG from Midjourney. Midjourney delivers JPEGs without embedded prompts, so a Discord download often carries no trace.
- Metadata can be edited. Its presence is strong evidence; its absence is no evidence at all.
How it works
The file is parsed byte by byte in your browser. For a PNG that means walking its chunk structure and reading every text chunk; for a JPEG, stepping through the segment markers to find EXIF, XMP and any JUMBF provenance box; for WebP, reading the RIFF containers.
Whatever is found is reported to you directly. There is no scoring model, no probability, and no pixel analysis — because a browser cannot do pixel-level AI detection honestly, and pretending otherwise is how these tools mislead people.
Nothing is uploaded. You can confirm that in your browser's Network tab.
Who uses this
Journalists verifying a submitted photo, moderators checking uploads, teachers reviewing coursework, buyers checking product images, and anyone who wants to know what a file actually says about itself.
Frequently asked questions
How accurate is this?
When it finds generation parameters or an IPTC AI tag, accuracy is not really the question — that data was written by the generator itself and is shown to you verbatim. When it finds nothing, it says inconclusive rather than guessing. There is no percentage because there is no model producing one.
Why does it say inconclusive for an obviously AI image?
Because the metadata was stripped. Every social platform removes it on upload, and so does taking a screenshot. The pixels may look artificial to you, but this tool reads the file rather than the picture — and it will not pretend otherwise.
Can it detect Midjourney or DALL·E images?
Only when the file still carries the evidence. DALL·E images from OpenAI include a C2PA manifest. Midjourney delivers JPEGs without embedded prompts, so a Discord download usually has nothing to find. Any tool claiming to reliably identify these from pixels alone is overstating what it does.
What is C2PA?
Content Credentials — a cryptographically signed record of how an image was made and edited, backed by Adobe, OpenAI, Google, Microsoft, Leica and Sony. This tool detects that a manifest is present; verifying its signature needs a certificate check, so use contentcredentials.org/verify to read it in full.
Is my image uploaded anywhere?
No. The file is parsed in your browser. Nothing is transmitted — which matters here, because people check sensitive images with tools like this.
Are my files uploaded to a server?
No. Everything runs inside your browser on your own device. Your file is never transmitted and never stored by us. You can confirm this by opening your browser’s Network tab while you use the tool — you will see no upload request.
Is there a limit on how many files I can process?
No. There is no daily cap, no account, and no watermark. Because the work happens on your device rather than our servers, there is nothing for us to ration.
Is this tool really free?
Yes — no account, no trial, no watermark, and no cap on usage. It runs on your own device rather than our servers, so there is no cost for us to pass on.
Does it work on mobile?
Yes. The interface adapts to phone and tablet screens, and everything runs in the browser, so there is no app to install.
Related tools
From other categories
Your files stay on your device. This tool runs entirely in your browser. No file is uploaded, transmitted, or stored. You can verify this by opening your browser's Network tab while using the tool.
Disclaimer. Specifications published by exam boards, governments and platforms change without notice. We verify and date every specification on this site, but you should always confirm against the official source before submitting an application. See our full disclaimer.