Skip to content
OpenDetect

Free · no sign-up · your file stays in your browser

Is this image AI? Check it here.

Drop in an image to see if it was made with AI. We look for the signals a file can't easily fake, show you each one, and tell you plainly when there isn't enough to go on.

What we check, in order

  1. Step 1

    Signed credentials

    If the file was made by ChatGPT, Gemini, Adobe Firefly and others, it often carries signed Content Credentials (C2PA). We read them on your device. A valid signature is very hard to fake.

  2. Step 2

    File metadata

    EXIF, XMP and IPTC fields can name the camera or the AI tool. Stable Diffusion and ComfyUI often save the prompt inside PNGs. Useful, but anyone can edit these.

  3. Step 3

    Invisible watermark

    If there are no credentials, we ask OpenAI's official checker to look for its invisible SynthID watermark, which survives screenshots and re-saving.

  4. Step 4

    Honest when it's unclear

    When none of this turns up anything, we say so. No signal is not proof of a real photo. We'll never show a made-up “97% AI” score.

How to read the result

  • Made with AIA signed credential or a watermark says AI. Strongest evidence there is.
  • Signs of AIUnsigned metadata names an AI tool. Usually right, but editable.
  • Signed camera photoSigned camera credentials and no AI signals.
  • Not enough evidenceNothing we can read either way. Common for screenshots and reposts.

What this can and can't tell you

Catches reliably

  • Original downloads from ChatGPT and GPT Image (credentials and watermark)
  • ChatGPT images that were screenshotted or re-saved (watermark)
  • Gemini, Adobe Firefly and other tools that sign their files
  • Stable Diffusion and ComfyUI PNGs that still hold their settings

Can't tell yet

  • Midjourney or Grok images after the metadata is gone
  • Gemini images after a screenshot (only Google can read that watermark)
  • Photos that were partly edited with AI and then re-exported
  • Anything older than the credentials and watermarks (before 2024–2026)

Questions

How accurate is this AI image detector?
We read the evidence the file carries: signed Content Credentials, metadata, and OpenAI's invisible watermark. When one of those says AI, that's close to certain. When none of them shows up, we can't tell, and we say so instead of guessing.
Why don't you give a percentage score?
Most other checkers run one AI model and print a percentage. Independent tests found those models often disagree with each other and flag real photos as AI. We show you what's actually in the file, so you can see why we reached our answer.
Do you keep my image?
No. Content Credentials and metadata are read in your browser. Only when we ask OpenAI's checker is the file sent to OpenAI, and we keep just a fingerprint of the file so the same image isn't checked twice.
Why does my screenshot say “not enough evidence”?
Screenshots, crops, and most social media uploads strip metadata. OpenAI's watermark often survives, but Midjourney, Grok and most open-source models don't add one. That's why “not enough evidence” is a common and honest answer.
Can it detect Gemini and Nano Banana images?
Yes, by its signed Content Credentials if the file is the original download. Google's invisible SynthID watermark can only be checked by Google for now, so we link you to Gemini's own check.
Is it free?
Yes, up to 10 OpenAI watermark checks a day for free. Reading credentials and metadata is unlimited because it runs on your device.

Other checkers