No single test can prove that an image was made with AI. What you can do is gather evidence. Some checks can show that an image was generated, while others only raise or lower your suspicion. It helps to start with the checks that give the firmest answers, then move to the weaker ones.
All of the methods below work in any browser, including Firefox. Firefox itself has no AI image checker, and its Page Info window (Ctrl+I, then the Media tab) only shows basics such as an image's address, type and size.
Start with the easiest check. Many sites now label AI content, or let people label their own:
Read the post's text and hashtags too. People often say "made with Midjourney" or tag a post #aiart. A label is useful evidence when it's there, but plenty of AI images don't have one, so a missing label tells you very little.
Content Credentials, built on a standard called C2PA, are signed records stored inside a file. They can say which tool made an image and whether AI was involved. OpenAI, Adobe, Google and Microsoft are among the companies that write them.
To check an image:
This is the strongest check on the list because the record is cryptographically signed, and editing the file after signing breaks the signature. It has one big limit: most social networks strip this data when an image is uploaded. If nothing shows up, the image may be a real photo, or it may be an AI image that has lost its credentials. Try to find the original file from the person or site that first published it.
Even without Content Credentials, a file can carry other clues. Some of the most useful are:
trainedAlgorithmicMedia means the file is tagged as generated by AI. compositeWithTrainedAlgorithmicMedia means a photo was edited using AI.The free command-line tool ExifTool shows all of this:
exiftool image.png
Online metadata viewers do the same job, but they usually upload your file to someone else's server, so avoid them for anything private.
Metadata works in one direction only. Finding a generator's settings is strong evidence of AI. Finding nothing proves nothing, since metadata is easy to remove and is usually stripped on upload. Camera details such as make, model and exposure point toward a real photo, but they can be faked too.
Downloaded AI images often keep their original names. Names that begin with Gemini_Generated_Image_, DALL·E or ChatGPT Image, or names like ComfyUI_00042_, are worth noticing. Anyone can rename a file, so treat this as a hint and look for something firmer.
Upload the image to TinEye or Google Lens to see where else it appears. This is the best method for images that claim to show a real event, because you're looking for the source rather than the technique:
If an image of a major event appears nowhere except a few social posts, be careful. That doesn't prove it's AI, but it does mean nobody has backed it up.
Older AI images were famous for extra fingers, and those mistakes still turn up. Zoom in and look for:
Current generators make far fewer of these mistakes, and real photos can have odd details too, especially after heavy compression. Use visual clues as a reason to dig further, not as a verdict.
AI image detection services, which judge the pixels directly, have a similar weakness. They give you a probability rather than an answer, and they can be wrong in both directions.
Here's a rough way to think about the results:
| Evidence | What it tells you |
|---|---|
| Content Credentials that declare AI | Very strong evidence of AI |
| Generator settings or an AI source tag in the file | Strong evidence of AI |
| A platform label or the creator's own statement | Good evidence of AI |
| An earliest copy on an AI gallery | Good evidence of AI |
| An earliest copy from a credible news agency | Good evidence that it's real |
| Visual oddities or a suspicious file name | A reason to keep checking |
| No labels, credentials or metadata | Almost nothing either way |
The last row matters most. Missing evidence is the normal state of most images online, whether they're real or generated.
Checking one image by hand is fine, but you can't do it for every image in a feed. AI Detector & Blocker is a free add-on for Firefox and Chrome from Terran Foundry that automates many of these checks as you browse. It looks for Content Credentials, metadata, generator settings, file names, platform labels and hashtags, and it runs on your device.
Images with evidence get an AI or AI? badge, and clicking the badge shows exactly what was found. It has the same limits as the manual checks: when a social network has stripped an image's data and nobody has labeled it, there may be nothing to find. A flag is a strong hint, not proof, and no flag doesn't mean an image is real. For anything important, a reverse image search is still worth the extra minute.