How to check if an image was generated by AI
Some AI images say so. Many do not. Absence of a label proves nothing.
In short
Look for what the image declares: an IPTC “digital source type” in its metadata, Content Credentials (C2PA), a visible watermark, or a label added by the platform. Any of these is a strong sign. But labels are easily lost — a screenshot or re-save removes metadata — so an image without them is not proven real.
Key takeaways
- A metadata label or C2PA manifest is strong evidence of AI involvement.
- No label is not evidence of a real photo.
- AI detectors that guess from pixels are unreliable.
1. Read the metadata
Generators that label their output write an IPTC field — DigitalSourceType — with the value “trainedAlgorithmicMedia”. Some local tools, such as Stable Diffusion interfaces, store the prompt in a PNG text field. The photo metadata tool on this site reads both.
2. Look for Content Credentials
Content Credentials (C2PA) are a signed record of how an image was made and edited. When present, they can say which tool generated or modified the image, and the signature shows whether the record was tampered with.
3. Visible signs and platform labels
Some generators add a visible mark in a corner. Social platforms add their own “AI info” or “made with AI” tags when they detect a declaration. Both are useful hints, and both disappear with a simple crop or screenshot.
What does not work well
- Online “AI detectors” that guess from pixels: they produce false positives and false negatives, and are easy to fool.
- Counting fingers or reading text: modern generators have largely fixed these tells.
When it matters
For an image that could mislead — news, evidence, a claim about a person — look for the original source, run a reverse image search, and check whether reliable outlets published it.
Photo metadata viewer and remover
Check what a photo gives away before you send it. Runs entirely in your browser.
Open the tool