How to Tell if an Image Is AI-Generated

A visible logo is the obvious tell, but plenty of AI images carry none — and still announce themselves in two places you can’t see by looking. Here are the three signals, from easiest to hardest to spot.

1. A visible watermark

The corner sparkle on Google Gemini images, or an on-image logo from other tools. Easy to see — and, because it sits on top of the picture, easy to remove cleanly.

2. Metadata inside the file (C2PA)

Most major tools now embed C2PA “Content Credentials”plus EXIF/XMP tags that record “made with AI.” You can’t see them on screen, but any tool that reads the file can. ChatGPT/DALL·E images usually have no visible logo but do carry this metadata.

3. An invisible watermark in the pixels (SynthID)

Google’s SynthIDweaves an imperceptible pattern across the pixels that survives crops and re-saves. You can’t detect it by eye, and only Google’s own detector reads it reliably — but if an image came from Gemini, Imagen or Veo, assume it’s there.

How to check any image

Drop it into Unmarkr and it scans the visible marks and the file’s metadata in seconds — and flags whether it likely came from a SynthID source. See our full AI watermark remover for what it can take off.