Guide

AI image detection: how to check if an image might be AI-generated

AI images are getting harder to spot with your eyes alone. A better approach is to check the file, provenance, editing signals, and the story around the image before deciding what to trust.

By Stipple Research12 min readUpdated 16 July 2026
Key takeaways
  • Do not rely on hands, teeth, or weird backgrounds alone. Visual clues change quickly.
  • Missing metadata does not prove an image is fake or real. It often just means the evidence is missing.
  • The best checks combine provenance, file evidence, visual review, and outside context.
  • Stipple can inspect uploaded images and documents for provenance, AI markers, and supporting signals.
Evidence path
  1. 01

    Save original

    Start with the material.

  2. 02

    Check provenance

    Add one more signal.

  3. 03

    Inspect file

    Add one more signal.

  4. 04

    Review image

    Add one more signal.

  5. 05

    Check context

    Make a careful call.

01

What is an AI image detector?

Short answer

An AI image detector is a tool that checks whether an image may have been generated or heavily changed by an AI system.

Some tools look for technical signals inside the file. Some look at pixels. Some look for signed provenance records, such as C2PA Content Credentials. Some use machine learning to compare the image with known AI-generated examples.

The tricky part is that an image can have many histories. It might be fully AI-generated, partly edited with an AI tool, compressed by a social platform, screenshotted, cropped, or created by a real camera and then changed later. A useful check should describe the evidence, not pretend every case is simply real or fake.

02

Why is AI image detection hard?

Early AI images often had obvious visual mistakes. Today, many do not. At the same time, normal editing and social media uploads can remove the very clues that would help you understand where an image came from.

  • New image models fix old visual tells quickly.
  • Screenshots create a new file and can wipe origin clues.
  • Social platforms often resize and recompress images.
  • A real photo can be edited with AI in only one small region.
  • A fake caption can mislead even when the image itself is real.
  • Metadata can be missing for ordinary reasons.
03

What should an image checker look at?

A strong review uses layers. One signal may be weak on its own, but several independent signals can give a much clearer picture.

Evidence layerWhat it can tell youWhat it cannot prove alone
Content CredentialsWhether a signed origin or edit history is attached.Whether the depicted event is true.
MetadataSoftware, timestamps, device clues, and sometimes generator details.Authenticity, because metadata can be removed or edited.
Pixel forensicsCompression, resizing, splicing, or other editing traces.The reason the edit happened.
AI classifierWhether the image resembles known AI-generated examples.Every generator or every future model.
Visual reviewOdd text, lighting, geometry, or object details.Proof, because real images can also look strange.
Context searchEarlier copies, source account, and external reports.Technical origin by itself.
04

What are C2PA and Content Credentials?

Short answer

C2PA is a standard for attaching signed information about where a digital file came from and how it was edited. Content Credentials are a common way people see that information.

When present and valid, provenance can be very helpful. It may tell you that an image was created by a camera, edited in a particular tool, or generated by software that recorded its role.

But missing credentials do not settle the question. A file may lose its credentials after a screenshot, export, crop, platform upload, or format conversion. So the right conclusion is often origin unknown, not definitely real or definitely fake.

05

How can you check whether an image is AI-generated?

Start by preserving the best version of the file. Do not work from a tiny screenshot if you can get the original upload, email attachment, or document.

StepWhat to do
Get the originalAsk for the highest-quality file, not a screenshot.
Check provenanceLook for signed credentials and metadata.
Inspect editsReview compression, cropping, splicing, and export traces.
Look at the imageCheck text, reflections, anatomy, lighting, and impossible details.
Check the storySearch for earlier copies, source accounts, and independent reporting.
Write down uncertaintySay what is known, what is likely, and what remains unresolved.
06

How should you read the result?

The safest language is evidence-based. Instead of saying this image is fake, say what the check found: valid AI provenance, missing origin metadata, editing traces, a classifier signal, or no strong signal.

FindingPlain-English interpretation
Valid AI provenance markerStrong evidence that the file records AI involvement.
No metadataThe origin is unresolved; this is common after upload or export.
Classifier signal onlyWorth reviewing with other evidence.
Editing traces onlyThe image changed, but the reason may be normal.
Clean file and credible sourceMore confidence, but not mathematical proof.
Conflicting signalsPreserve the file and review more carefully.
07

What can Stipple inspect?

Stipple can inspect uploaded images and documents for explicit AI-generation markers, C2PA/IPTC-style provenance, generator metadata, editing traces, and other supporting signals.

The goal is not to guess dramatically. The goal is to give you a careful, explainable result that says what the file evidence supports and what it does not.

Questions

Frequently asked questions

Can an AI image detector be 100% accurate?

No. AI image generation and editing tools change quickly, and normal file transformations can remove useful clues.

Does missing metadata mean an image is fake?

No. Missing metadata is common. It means one evidence channel is absent, not that the image is fake or real.

Can Stipple check AI-generated images?

Yes. Stipple can inspect uploaded images and documents for provenance, AI markers, editing traces, and related signals.

Sources

Sources and further reading

  1. 01C2PA technical specification
  2. 02Content Authenticity Initiative
  3. 03NIST Open Media Forensics Challenge

Educational guidance, not a forensic certification. Detection technologies and standards change; review material decisions against current evidence.

Inspect the file, not just the pixels

Upload an image or document to review provenance, synthetic indicators, editing traces, and other evidence in one place.

Verify an image