Which AI Made This Image?

Sometimes the image tells you: a mark in the corner, signed Content Credentials, a telltale size. Here is how to check each one, what every major generator leaves behind, and when the honest answer is that you cannot know.

โœ“ 100% freeโœ“ No sign-up ๐Ÿ”’ Deleted after analysisโšก Instant
โฌ†๏ธ

Drag & drop, paste, or click to upload

PNG ยท JPG ยท WEBP ยท HEIC โ€” your image is analyzed privately and deleted

No image? Try one:

Five checks, in order

  1. Look at the corners. Some generators stamp a visible mark: a Gemini sparkle on free-tier Google images, a small “Grok” mark on some Grok images, a moving cloud logo on Sora video downloads. A mark can be cropped out, so its absence proves nothing.
  2. Read the Content Credentials. OpenAI, Google and Adobe sign their images with C2PA Content Credentials that name the tool. The metadata checker reads them from the original file. Screenshots and most re-saves remove them.
  3. Read the rest of the metadata. Some Stable Diffusion interfaces write the prompt and the model name into the PNG file itself. Worth a look, though easily removed.
  4. Check the size. Generators favour particular dimensions: exact squares such as 1024×1024 or 2048×2048 on Google’s models, multiples of 64 near one megapixel on SDXL, about one megapixel on SD 3.5. A size is a clue, never proof, and it only rules out an unedited phone photo.
  5. If nothing is left, you usually cannot tell which AI made it. A screenshot of a re-shared image has lost every trace above. You can still ask whether any AI made it: the check above estimates that from the pixels alone.

What each generator leaves behind

GeneratorVisible markContent Credentials (C2PA)Invisible watermark
ChatGPT / GPT Image (OpenAI)None on stillsYes, signed by OpenAISynthID since 19 May 2026
Gemini, Nano Banana, Imagen (Google)A Gemini sparkle in a corner on free and Pro-tier imagesYesSynthID, always
Adobe FireflyNoneYes, signed by AdobeA durable Content Credentials signal
MidjourneyNoneNoNone documented
Grok (xAI)Sometimes a small “Grok” mark in a cornerNoNone documented
Flux / Flux 2 (Black Forest Labs)NoneOnly from the paid API (Flux 2)Not on images made locally
Stable Diffusion, SDXL, SD 3.5NoneNoSometimes on SD 1.5 and SDXL; not on SD 3.5
Sora (OpenAI video)A moving cloud logo on app downloadsIn the video file only, not in frames or screenshotsNo, on the evidence available

Each name links to its own guide, with the sources and the visual tells for that generator. Invisible watermarks such as Google’s SynthID are not readable by third-party tools, so they cannot be checked here.

What our check can and cannot tell you

The detector above estimates whether an image was made by AI. It does not name the generator. To name it, you need one of the traces in the checks above, and the metadata checker is the fastest way to read them.

If you only need to know whether a picture is real, the AI image tester is the quicker page.

Frequently asked questions

How can I tell which AI made an image?

Check, in order: a visible mark in a corner, the Content Credentials in the file, any other metadata, and the image size. If the file is a screenshot or has been re-saved, those traces are usually gone and the generator cannot be identified.

Can a detector tell which AI generator made an image?

Ours cannot. It estimates whether an image was made by AI at all. Naming the generator needs a trace in the file or the picture, such as signed Content Credentials or a visible mark.

Which AI generators sign their images?

OpenAI, Google and Adobe attach C2PA Content Credentials that name the tool. Midjourney, Grok and open Stable Diffusion models do not.

Does a missing watermark mean the image is real?

No. Marks can be cropped out, metadata disappears when an image is re-saved, and several generators add neither. Use the AI check for the whether question.