Check whether an image may be AI-generated
Look for readable generator metadata, C2PA claims, and other file-level evidence before drawing a conclusion.
Free AI Image Detector
Upload a file to inspect AI-generation traces, C2PA Content Credentials, metadata, camera clues, and editing history. See what the file can support—and what it cannot prove.
Waiting for upload
{count} chars
Check result
After upload, the system will provide a reference assessment based on clues in the image.
Was this tool helpful?
{}
Start with the right question
Use ImgShield to check whether an image may be AI-generated, then read the evidence separately: generation records, C2PA claims, camera metadata, editing traces, and missing information. A missing record is not proof that a picture is AI-made.
Look for readable generator metadata, C2PA claims, and other file-level evidence before drawing a conclusion.
Useful for analyzing articles, emails, comments, or product copy for common AI-writing clues.
Images are analyzed locally first, and text checks stay in your browser, making it useful for quick screening and shareable reports.
How it works
After upload, the tool reads visible provenance claims, generation metadata, capture fields, and editing traces in the file. Those clues have different strength, so the result explains what they support and what is missing; one score cannot establish an image's true origin.
Checks for common generation clues such as Stable Diffusion, Midjourney, ComfyUI, Prompt, and Seed.
Checks for phone or camera model, aperture, shutter speed, ISO, focal length, and capture time.
Parses C2PA/JUMBF provenance records, PNG/JPEG/WebP metadata, and hints left by tools such as Photoshop or Lightroom.
Looks for templated phrasing, overly regular structure, repeated phrases, and missing concrete detail.
AI image detection guides
A score is a starting point. Learn when to check the source, inspect the original file, and ask for human review.
Review the original file, metadata, C2PA records, visual clues, source context, and reverse image search results.
Read the image evidence guideOrganize triage, evidence records, and a second review. Keep uncertain cases visible instead of relying on one score.
Read the bulk review workflowUnderstand metadata, provenance, generator traces, model signals, and writing patterns.
Read the methodologyPrepare original files, inspect generation parameters, and understand the limits of screenshots and compressed images.
Read the image detection guideLearn about structure, repetition, specific details, and why short passages are difficult to assess.
Read the text detection guidePlans
Local metadata and provenance checks remain free. A plan adds monthly Sightengine image-model checks for uncertain results. You can request cancellation before the next billing period.
$9.90
per month
400 enhanced checks / month
Recommended
$29.90
per month
1,150 enhanced checks / month
$59.50
per month
2,250 enhanced checks / month
Waffo processes payments. Your allowance is updated after payment is confirmed; returning to this page alone does not confirm a payment.
Privacy
Text checking stays in this browser. If a local image result is uncertain, the current image is sent through the same-origin backend to Sightengine for an additional pixel-model signal. See the privacy notice for data use and cookie choices.
Important note
AI image and text detection have no absolute answer. Screenshots, compression, reposting, and edits can remove image clues; text can be rewritten by a person, polished by AI, or deliberately mimic another style. For important cases, combine this result with source checks, account context, and human review.
FAQ
No. It gives a reference judgment based on locally readable clues, but it cannot guarantee absolute accuracy.
If an image has been compressed, screenshotted, or stripped of generation data, the page has fewer clues and gives a more cautious result.
Yes. Reposted, edited, or screenshotted photos may lose camera metadata, so the tool uses cautious wording.
If the local check gives a clear direction, the image stays in the current browser tab. If the local result is uncertain, the current image is sent through the backend to a third-party detector for an additional signal.
No. It only gives a reference risk based on writing structure, tone, repetition, and detail. Short text is especially uncertain.