Marked as AI by a tool
A trainedAlgorithmicMedia source type or a prompt field means the creating app labelled the file as generated.
Get an answer from the evidence inside the file: Content Credentials, AI tags and saved prompts. It does not guess from pixels.
Inspect supported file records without editing the original.
Start with what the file says about itself.
The most reliable first check is the file. Generators like ChatGPT and Adobe Firefly sign their images with Content Credentials, Meta AI tags them with IPTC metadata, and Stable Diffusion saves the prompt.
This checker reads that evidence in seconds and shows it as plain findings: a generation field, a saved prompt, or a credential container.
It does not guess from pixels. Visual AI detectors give a probability that can be wrong either way. File evidence is a fact, but it can be stripped, so an empty result is not proof the image is real.
Many AI images have their metadata stripped. Other ways to check:
Three steps, about ten seconds per image.
Use the original file if you can. A screenshot or social media download has usually lost its metadata.
Look for a decoded generation field, a recognized prompt record or a provenance container.
A signed credential is strong. Unsigned tags are a clue. Nothing found means unknown, not real.
Two different ways to answer the same question.
| This checker (file evidence) | Visual AI detectors | |
|---|---|---|
| What it reads | Records stored in the file: C2PA, IPTC, XMP, PNG text. | The pixels, using a trained classifier. |
| What you get | A finding you can see and verify, with the field name. | A probability score, such as 87% AI. |
| When it fails | Metadata was stripped by a screenshot, re-save or upload. | Edited, compressed or new-model images fool the score. |
| Privacy | Runs in your browser. The image is not uploaded. | Usually uploads the image to a server. |
For the strongest answer, use both: file evidence first, then a visual detector if the file has no records.
JPG, PNG, WebP, AVIF and HEIC still images. Up to 30 files, 15 MB each. Images up to 50 megapixels; animated PNG and WebP are not supported. Free, no account.
Read the report as evidence, not a verdict.
A trainedAlgorithmicMedia source type or a prompt field means the creating app labelled the file as generated.
PNG text named parameters, prompt or workflow is what Automatic1111 and ComfyUI save. It often contains the full prompt.
A C2PA container records how the image was made. Check its signer in the C2PA checker.
No records does not mean real. Most platforms strip metadata, and so does a screenshot.
Answers about results, privacy and limits.
When the file has Content Credentials, the answer comes from a signed record, not a guess. When it has nothing, the result is 'no AI record', which is not the same as 'human-made'.
The metadata was probably stripped by a screenshot, an editor or a social media download.
Yes, if it's the original download. ChatGPT images carry Content Credentials signed by OpenAI.
No. Percentage scores come from pixel classifiers and can be wrong either way. This shows the evidence in the file.
Not a visual one. It checks the evidence stored inside the file. It does not score pixels, so it never guesses.
You can, but most social platforms strip metadata on upload, so the result will often be empty. Ask for the original file when it matters.
Unsigned fields, yes, with any metadata editor. A C2PA credential is signed, so tampering shows up as an invalid signature in the C2PA checker.
Only if the file still carries Midjourney’s metadata. What Midjourney writes has changed over time, so check the decoded fields in the report.
Yes. It is free, with no account, credits or trial. The file and batch limits are shown on the upload box.
No. Files are read in your browser. Save your results before you clear the batch or leave the page.
Free, private and instant. No signup.