AI image detector, free and private
Drop an image to see if it was made with AI. The tool reads the labels in the file, looks for a hidden watermark and scores the pixels. The image never leaves your computer.
Drop an image, or paste one
JPEG, PNG, WebP, AVIF, HEIC
The first check downloads about 44 MB, once.
How it works
- 1Add the image
Drop it, choose it or paste it. The original file keeps more clues than a screenshot.
- 2Read the labels
Most big AI tools write a label into the file. Google, OpenAI and Adobe sign theirs, and the tool checks the signature.
- 3Check the pixels
A model trained on images from thousands of AI tools scores the pixels. A decoder looks for the hidden Stable Diffusion watermark.
How far to trust the result
A signed label is the best evidence there is. When its signature checks out, a known company signed those exact pixels.
The pixel score is an educated guess. In our test it caught about half of the AI images, and fewer of the newest ones. It rarely flags a real photo, but a low score does not mean the image is real.
For a better check
- Use the original file. A screenshot or a social media upload strips the labels and blurs the clues in the pixels.
- Check the largest copy you can find. Resizing erases the Stable Diffusion watermark.
- Look closely yourself. Text on signs, hands and jewellery still trip up many AI tools.
- For images that may come from Google or ChatGPT, also try Google's checker at synthid.com.
Made by the team behind Cedric
Cedric is an AI teammate in Slack. He joins your Google Meet, Zoom and Teams calls, posts the minutes when the call ends and follows up on every to-do.
FAQ
Yes. There is no account and no daily limit. Your own device runs the check, so an image costs us nothing.
No. The file stays in this browser tab. The page downloads the checker once, about 44 MB, and never sends your image anywhere.
Start with the labels. Many AI tools write one into the file, and Google, OpenAI and Adobe sign theirs. When the labels are gone, look at the pixels: this tool's score, then details like text and hands. No single check catches every AI image.
Yes, while the file keeps its Content Credentials, which both add and sign. A screenshot or a social media upload removes them. After that only the pixel score is left, and it misses many images from the newest tools.
A mark hidden in the pixels that stays when the file's labels are stripped. Google's mark, which OpenAI also uses now, can only be read by Google's own checker at synthid.com. The official Stable Diffusion code adds an open mark, and this tool reads it. Resizing or heavy compression erases that one.
A signed record inside the file that says which tool made the image and what was done to it. Google, OpenAI and Adobe add them, and so do some cameras and phones. When the signature checks out, nobody changed the pixels after signing.
We tested it in October 2026 on 754 images, 448 from current AI tools and 306 real photos. At a score of 90% or more it caught 42% of the AI images and wrongly flagged 2 real photos. From 50% it caught 56% and flagged 11. A low score proves little.
Heavy filters, upscaling apps and phone beauty modes leave patterns that look like AI. Treat a high score as a reason to look closer, and trust a signed label over the score.
Your AI employee is waiting for you
Get started for free100K credits ($50) on sign up. No card needed.