ExifZero What this checker looks at
AI image tools leave three kinds of traces in a file. This checker reads all of them and explains what each one means, so you can judge the image yourself instead of trusting a single score.
1. Content Credentials (C2PA)
A signed manifest that records who created the image and what was done to it. OpenAI, Adobe, Google, Microsoft and several camera makers write one.
When the manifest's action says the digital source type is trainedAlgorithmicMedia, the creator declared the image AI-generated.
2. IPTC Digital Source Type
A standard XMP field used by news agencies and stock sites, and by Midjourney and Meta AI. The value
trainedAlgorithmicMedia means “created by generative AI”; compositeWithTrainedAlgorithmicMedia means a real image was edited with
generative AI.
3. Generation parameters
Local tools such as Stable Diffusion WebUI and ComfyUI save the prompt, seed, sampler and model into the PNG so the image can be recreated. When we find them, we show the prompt in plain text.
How to read a “No AI tags found” result
It only means the metadata carries no AI marks. Screenshots, re-saved copies and images downloaded from social networks usually have no metadata at all. For those, look for visual clues, reverse-image search the picture, or ask the source.
Frequently asked questions
Can this tell whether any image is AI-generated?
It tells you what the file says about itself. If an AI tool left a label, a C2PA manifest or its generation settings, you will see them. If the metadata was removed — as most social networks do on upload — there is nothing left to read, and the result is “No AI tags found”. That is not proof the image is real.
Which AI generators can it recognise?
Over 30, including ChatGPT / DALL·E, Midjourney, Adobe Firefly, Google Gemini and Imagen, Stable Diffusion (A1111, Forge, SD.Next), ComfyUI, FLUX, NovelAI, InvokeAI, Fooocus, Leonardo, Ideogram, Microsoft Designer, Meta AI, Runway, Krea, Recraft, Grok, Seedream / Jimeng, Tongyi Wanxiang and Hunyuan.
What does “Strong”, “Medium” and “Info” mean?
Strong signals are explicit: an IPTC “trainedAlgorithmicMedia” label or embedded generation parameters. Medium signals mention an AI tool or algorithmic creation. Info items give context, such as the presence of Content Credentials or camera exposure data.
Can metadata be faked?
Yes. Plain EXIF and XMP fields can be edited by anyone. C2PA manifests are signed, but this page displays the signer without verifying the signature. Treat every result as evidence, not a verdict.
Does it detect pixel watermarks such as SynthID?
No. Pixel watermarks need the vendor’s own detector. Google offers a SynthID check in Gemini for images made with its tools.