AI photo checker

What it looks for

Checkobot detects AI-generated and AI-manipulated images, with a dedicated face-forensics model for deepfakes. The whole-image analysis looks at every picture, whatever's in it: people, places, products, food, animals or art. When there's a face, a second detector looks at the face on its own.

Photos people check most

• Profile photos on dating apps and social networks. If the face analysis flags it, ask for a fresh photo or a video call before you trust it.
• Listings: a flat, a car or a product that looks better than it should. Check every photo in the listing, not just the first.
• Reviews and before-and-after shots, where an AI edit can quietly do the selling.
• News and viral photos. Check the largest copy you can find, and look for who posted it first.
• Photos of people you know, sent by someone you don't. A swapped face usually belongs to someone who never agreed to it.

Reading the result

• AI: at least one detector crossed its threshold. The result says which one: face analysis, whole-image analysis, or both.
• No AI detected: neither detector found signs at its threshold. That isn't proof the photo is untouched, only that these two detectors didn't flag it.
• Refused: the image was too small or too compressed to judge fairly. Find a bigger copy; a refused check doesn't use up a free one.
• Every result shows the model and rules version it was checked with, so a result can always be traced back to how it was made.

Where it struggles

• Talking-head avatars and lip-sync from new engines are our weakest area: it misses many of them.
• Illustrated, cartoon, CGI or doll faces can trip the face analysis. It learned from photographs, so drawn faces can look odd to it.
• Very low-quality or heavily distorted photos get flagged by mistake more often than clean ones.
• Documents, IDs and receipts aren't what it's built for, and it doesn't check audio or liveness (printed photos held up to a camera, screen replays).

Questions