How To Detect AI Generated Images Online

Online AI image checkers fall into three groups, and each tests something different. Knowing which is which keeps you from over-trusting a single result.

1. Provenance checkers

These read records attached to the file. Content Credentials Verify reads C2PA Content Credentials, a signed history added by tools such as OpenAI's image generators, Adobe Firefly and some cameras. It tells you who signed the record and whether it's intact.

• Reliable when present: a valid record saying AI made the image is strong evidence.
• Often absent: screenshots and social media strip it.

2. Metadata readers

These show the hidden fields in a file: camera details, software, AI tool names and settings. In Chrome, AI Image Detector (made by Patrick Bushe, who runs this site) reads them when you right-click an image, without uploading it, and gives a verdict with the evidence. Online EXIF viewers do the same for uploaded files.

• Good at: catching AI tool names and Stable Diffusion settings in original files.
• Weak when: metadata has been stripped or edited.

3. Visual detectors

These analyze the pixels with a trained model and return a probability. Hive, SightEngine, Illuminarty and AI or Not offer free tiers or demos.

• Good at: images with no metadata.
• Weak at: heavily compressed, cropped or edited images, and images from generators newer than the model's training.
• Results vary: two detectors can disagree on the same image.

Don't forget reverse search

Google Lens, TinEye and Bing aren't AI detectors, but finding an image's earliest copy often answers the question faster than any of them.

Reading the results

• A provenance record or AI metadata: strong evidence.
• A high visual score alone: a reason to look closer.
• No metadata: tells you nothing either way.
• Conflicting results: treat the image as unverified.

See free tools to check an image and avoiding false positives.