100% freeNo signupNo upload

ChatGPT Image Detector

Was this picture made with ChatGPT or DALL·E? Drop it in to look for OpenAI's signed Content Credentials and the traces AI images leave behind.

Everything runs in your browser. Your image is not uploaded.

How ChatGPT marks its images

OpenAI signs the images ChatGPT makes. Each file carries a C2PA manifest, the same Content Credentials standard used by Adobe and a growing number of camera makers. The manifest says the image was created by an OpenAI model and is signed with OpenAI's certificate, so it can't be forged or moved to another image without the signature failing.

That makes ChatGPT images among the easiest AI images to identify, as long as you have the original file. A valid OpenAI signature is about as close to proof as image checking gets.

How to check a ChatGPT image

  1. Get the original file. Download the image from ChatGPT, or ask whoever sent it for the file itself rather than a screenshot or a forwarded copy.
  2. Drop it into the checker above. Sens reads the Content Credentials, scans the metadata and runs the visual detector at the same time.
  3. Read the verdict.“AI-generated” with OpenAI named as the signer settles it. Anything weaker comes with the evidence and caveats that explain why.

When the credentials are gone

Content Credentials live in the file, not the pixels. They disappear when someone takes a screenshot, sends the image over WhatsApp or another messenger, posts it to most social networks, or re-saves it in an editor that doesn't keep them. Most ChatGPT images you meet in the wild have been through at least one of those.

Without a manifest, Sens looks for other traces (OpenAI and ChatGPT tags in XMP or EXIF survive some re-saves) and then relies on its neural detector, which was trained on output from thousands of image generators. The detector alone can say “Likely AI-generated” but never “AI-generated”, and it can't tell you which generator made the image. Reading the verdict explains each headline.

Not only ChatGPT

The same check covers Adobe Firefly, which also signs its output, Stable Diffusion tools that write their prompts into PNG files, and images from Midjourney, Gemini, Flux and others through the visual detector. The AI image detector explains how each signal is weighed, and the C2PA checker shows the full manifest when you want every detail. For text, the AI text detector checks whether a passage reads like ChatGPT wrote it.

Frequently asked questions

Can you tell if an image was made by ChatGPT?
Often, if you have the original file. OpenAI adds signed C2PA Content Credentials to images made with ChatGPT and DALL·E, and Sens verifies that signature in your browser. When the signature is intact, the result names OpenAI as the signer and the verdict is AI-generated. If the credentials have been stripped, the answer falls back to metadata and a visual detector, which gives a probability rather than proof.
Does ChatGPT put a watermark on its images?
Not a visible one. The marker is a C2PA manifest embedded in the file: a signed record saying which tool made the image. It does not change the pixels, so anything that rewrites the file without copying the manifest, such as a screenshot or an upload to most social networks, removes it.
I checked an image I made in ChatGPT and Sens found no AI signs. Why?
The file you checked is probably not the one ChatGPT produced. Screenshots, messaging apps like WhatsApp, and social platforms re-encode images and drop the Content Credentials. Download the image from ChatGPT directly and check that file. Without credentials, the visual detector still runs, but a phone-friendly re-compressed copy gives it less to work with.
Does it work for DALL·E 3 and GPT-4o images?
Yes. Images from ChatGPT's image generation, DALL·E 3 and OpenAI's image API are signed the same way, so the check is the same for all of them.
Can a real photo be signed by OpenAI?
Only if it went through ChatGPT. Photos you upload to ChatGPT and ask it to edit come back as new images signed by OpenAI. Sens reports what the manifest says, so the edited copy shows up as made by OpenAI, which is accurate: the pixels you are looking at came out of the model. The photo you started with is unaffected.
Is my image uploaded?
No. The Content Credentials check, the metadata reader and the neural detector all run in your browser. The image only leaves your device if you sign in and choose to save a report.

More free checks