SynthID Detector

Guide · AI image checks

How to tell if an image is AI-generated

There is no single test that tells you whether an image was made by AI. There are, however, several checks that each answer part of the question, and together they usually give a clear picture. Start with the evidence the file carries about itself, then use the official tools from the companies that made the most common AI images, and only then fall back on looking for visual clues.

Check an image

In short

  • Signed Content Credentials (C2PA) are the strongest evidence when a file still has them.
  • Google's and OpenAI's own checkers can find their SynthID watermarks even after metadata is gone.
  • A missing watermark or missing credentials does not prove an image is real.
  • Visual clues and the image's source are supporting evidence, not proof.

1. Read the Content Credentials

Content Credentials are a C2PA manifest: a block of signed data stored with the image that records how it was made and edited. The C2PA describes them as tamper-evident and cryptographically signed, with hashes of both the image and the provenance data, so any later change breaks the link. Google, OpenAI, Adobe and other tools write them into the images they generate.

Upload the original file to the SynthID Detector on this site, to the official C2PA verify tool or to another Content Credentials reader. If a trusted manifest says the image was created by a generative AI tool, that is strong evidence. If the manifest is present but no longer matches the image, treat every claim in it as unverified.

2. Look at the metadata

Many generators also write plain metadata: an IPTC digital source type that marks AI-generated media, a software name in EXIF or XMP, or the prompt and settings in a PNG text chunk. These fields are useful hints, but unlike Content Credentials they are not signed, so anyone can add, change or remove them. The free check on this site scans EXIF, XMP and PNG text for these traces.

Screenshots, many messaging apps and most social networks remove both Content Credentials and metadata, so always try to find the original upload first.

3. Test for a SynthID watermark with the official tools

SynthID is an invisible watermark from Google DeepMind that sits inside the pixels and is designed to survive edits that strip metadata. Google says it has watermarked over 100 billion images and videos, and that companies such as OpenAI, Kakao and ElevenLabs are bringing SynthID to their AI content. Only the companies' own tools read it:

  • Google: upload the image to the Gemini app and ask 'Is this AI-generated?' or 'Was this created or edited by Google AI?'. Gemini's help page says it can currently recognise only content created by Google AI tools.
  • Google Search and Chrome: Google announced on May 19, 2026 that SynthID verification is expanding to Search features such as Lens, AI Mode and Circle to Search, and to Chrome.
  • OpenAI: upload the image at openai.com/verify, which looks for an OpenAI SynthID watermark or a trusted OpenAI C2PA manifest.

4. Look closely, but do not rely on it

Older or cheaper generators still leave visible traces: hands with the wrong number of fingers, text that turns into nonsense, jewellery that merges into skin, inconsistent reflections and shadows, or repeated patterns in backgrounds. Newer models make far fewer of these mistakes, and real photos can look strange too. Use what you see to decide which other checks to run, not as a verdict.

An AI visual analysis, such as the optional one on this site, does the same thing more systematically and explains what it finds. It is an informed estimate, useful when every other trace has been removed, but never proof on its own.

5. Trace where the image came from

A reverse image search can show the earliest place an image appeared, whether a news agency or a known photographer published it, and whether it has been altered from an earlier version. An account that posts many similar images, or a story that only one unknown source carries, deserves more caution. Combine this with the technical checks above before you share or rely on an image.

Reading the results together

A trusted C2PA manifest that names an AI generator, or a positive result from Google's or OpenAI's checker, is strong evidence that the image is AI-generated or AI-edited. A negative result is weaker: Gemini's help page notes that no SynthID watermark means the image was not made by Google AI but could have been made by other AI systems, and OpenAI's developer documentation says a not-detected result does not rule out AI content either. When the checks disagree or find nothing, say what you know and what you do not.

Frequently asked questions

What is the most reliable way to tell if an image is AI-generated?

Signed Content Credentials from the generating tool, or a positive result from the tool maker's own checker (the Gemini app for Google AI, openai.com/verify for OpenAI). Both are far more reliable than looking for visual mistakes.

Can an AI image have no watermark at all?

Yes. Only some companies embed SynthID or Content Credentials, and screenshots and social platforms remove metadata. Many open-source models add no watermark at all.

Does this site read the SynthID watermark?

No. It reads Content Credentials and metadata, including whether a SynthID watermark is declared in them. To test the pixel watermark itself, use the Gemini app or openai.com/verify.

Can a tool prove an image is real?

No. Camera-signed Content Credentials, such as those from Pixel 10, show that a photo was captured by a camera and how it changed, but a missing AI signal on its own proves nothing.

Sources

Updated 2026-10-29. SynthID Detector is an independent service and is not affiliated with Google, Google DeepMind or OpenAI. SynthID is a Google technology. Results describe the provenance data found in a file and are not legal proof of how an image was made.