AI Image Detector
Reads the provenance records inside an image file to tell you whether it was made with AI or taken with a camera. Checks C2PA Content Credentials from ChatGPT, Google and Adobe, the "Made with Google AI" label and Stable Diffusion settings.
🔒 Your images are not uploaded — they are checked only on this device (in your browser).
What does it read?
| Record | What it is | Who adds it |
|---|---|---|
| C2PA Content Credentials | A provenance record of when and with what the image was created and edited, digitally signed by the company | OpenAI (ChatGPT images), Google, Adobe, Microsoft, some cameras and smartphones |
| IPTC Digital Source Type | A standard label such as "AI-generated image" or "photo taken with a camera" | Google ("Made with Google AI"), Adobe, photo management software |
| PNG text chunks | The prompt and settings used to generate the image | Stable Diffusion (WebUI), ComfyUI and some AI services |
| EXIF camera data | Make, model, capture time and the software that saved the file | Cameras and smartphones |
How far can you trust the result?
- "There is a record that this was made with AI" — fairly reliable. These records are added automatically by the companies, and people rarely add an AI label on purpose.
- "There is a record that this was taken with a camera" — treat it as a hint only. Anyone can edit EXIF data. It is more trustworthy if a C2PA signature is present too.
- "No provenance records" — this tells you nothing either way. Records disappear when an image passes through messaging apps, social media or a screenshot.
When there is no record, you can check for a hidden watermark with Google's SynthID Detector (Google, OpenAI, NVIDIA and Kakao models, sign-in required). Digital signatures can be verified on the official C2PA verification site. Want to blur people before sharing an image? Try blur faces. More tools are in the image tools collection.
Frequently asked questions
If there is no record, does that mean a human took the photo?
No. Uploading an image to social media or a messaging app, or taking a screenshot of it, often strips the provenance data — and it can also be removed on purpose. No record means "can't tell", not proof that a person made it.
Which AI image generators leave a record?
As of 2026, major companies including OpenAI (ChatGPT), Google Gemini, Adobe Firefly and Microsoft embed C2PA provenance data (Content Credentials) in the images they generate. Google images also carry a "Made with Google AI" label and a record that a SynthID watermark was added. PNGs made with Stable Diffusion or ComfyUI often contain the generation settings as text.
Does it verify that the record is genuine?
This tool reads and displays the records inside the file; it does not check whether the digital signature has been tampered with. To verify the signature, drop the same file into the official C2PA verification site (verify.contentauthenticity.org).
Can it detect SynthID watermarks?
Not directly. SynthID is hidden in the image pixels and can only be checked with Google's own detector. On 7 October 2026 Google opened its SynthID Detector (synthid.com) to the public; it checks images, video and audio from Google, OpenAI, NVIDIA and Kakao models (sign-in required). CalcLab tells you when the file contains a record saying SynthID was applied.
Are my images uploaded to a server?
No. The records are read directly in your browser. You can check several images at once.