OpenAI GPT Image

ChatGPT image detector

92.1–93.9%of ChatGPT images caught in testing

Paste an image or a link anywhere (Ctrl/⌘ + V)

Your first scan is free. By scanning, you accept our Terms. Images and results are kept until you delete them (Privacy).

Fewer than 1 in 100 original camera photos are wrongly flagged (0.6%)

In short

Can Perspicax detect ChatGPT images?

Yes. Perspicax caught 92.1–93.9% of ChatGPT images in testing, across GPT Image 2 and GPT Image 2.5. ChatGPT's files carry Content Credentials and watermarks at first, but uploads to X and most social apps strip the label, so Perspicax also reads the pixels: on 10,072 GPT Image 2 images posted on X, it caught 88.3%.

Where these images come from

OpenAI's GPT Image models make the pictures in ChatGPT. Developers use them through OpenAI's Images API, and apps such as Canva and Adobe Firefly offer them too.

Labels and watermarks

At the source, OpenAI adds Content Credentials (C2PA), its own invisible watermark and, since 2026, Google's SynthID. Labels in the file don't survive uploads to X and most social apps: in our sample of GPT Image 2 images collected from X, none still had Content Credentials.

Content Credentials (C2PA)Live
SynthID watermark checkSupported

What we measured

GPT Image 2.5 Sunburst93.9%
GPT Image 2.5 Flare92.6%
GPT Image 292.1%

Worth knowing

GPT Image is strong at layouts with text, such as posters and menus. Perspicax checks photo-like images; graphics and screenshots come back as outside scope.

How it works

Questions about ChatGPT images

Does ChatGPT label the images it makes?

Yes, at the source: OpenAI attaches Content Credentials (C2PA) and invisible watermarks. The C2PA label is lost when the image is re-saved, screenshotted or uploaded to most social apps.

Can Perspicax check ChatGPT images posted on X?

Yes. On 10,072 GPT Image 2 images posted on X, where the platform had stripped Content Credentials, Perspicax caught 88.3%.

Can it tell GPT Image 2.5 from GPT Image 2?

No. It tells you whether a photo was fully generated by AI, not which model made it. We measure each model separately: GPT Image 2.5 Sunburst 93.9%, GPT Image 2.5 Flare 92.6% and GPT Image 2 92.1%.

Does it work on GPT Image photos made in Canva or Firefly?

Yes. Perspicax checks the image itself, whichever app made it. Labels differ from app to app, so the pixel check matters most.