Free · no sign-up · your file stays in your browser
Is this image AI?
Ask the file.
Got one picture you can't make your mind up about? Drop it in. You get a straight answer when the file has proof, and an honest “can't tell” when it doesn't.
Same AI picture, two copies.
Run both. One still has its signed credentials, the other only has the hidden watermark left.
No file handy? Try one:
What we check, in order
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.
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.
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.
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.
The answer depends on where you got it
The same AI picture can be easy or impossible to check, depending on which copy you have. Here's what to expect.
- Downloaded straight from an AI toolUsually a clear answer. ChatGPT, Gemini and Adobe Firefly sign the files they make.
- Saved from Instagram, X, Facebook or RedditOften “not enough evidence”. Most apps strip the metadata on upload. OpenAI's watermark can still turn up, so it's worth a try.
- A screenshot someone sent youWe'll tell you it's a screenshot. The original's credentials are gone, but OpenAI's watermark often survives. Ask for the original file if you can.
- An old photo or a scanProbably no answer. Credentials and watermarks only started showing up in 2024, so older files have nothing to read.
Checking a viral photo or a meme? The real or fake check adds an error-level map and reverse image search. Checking lots of images? The AI image detector shows every signal it reads, with samples to compare.
Questions
- Can you just tell me yes or no?
- When the file proves it, yes: a signed credential or OpenAI's watermark gives a clear “Made with AI”. When the file has nothing we can read, you get “not enough evidence” instead of a coin-flip guess dressed up as a score.
- What's the difference between “Made with AI” and “Signs of AI”?
- “Made with AI” means a signed credential or a hidden watermark says so, which is very hard to fake. “Signs of AI” means plain metadata names an AI tool. That's usually right, but anyone can edit metadata, so treat it as a strong hint.
- Will it catch a Midjourney or Grok image?
- Only if it's the original file and the metadata is still there. Midjourney and Grok don't add a watermark we can read, so once the image has been reposted or screenshotted there's nothing left to check. Pixel-guessing detectors will still give you a number, but independent tests keep finding they disagree with each other.
- Does it work on my phone?
- Yes. It runs in your phone's browser. Tap the box to pick a photo from your gallery, or paste a screenshot.
- Is it free?
- Yes. No sign-up, and up to 10 OpenAI watermark checks a day. Reading credentials and metadata is unlimited because it happens on your device.