AI Image Checker
Check a photo for AI provenance marks: C2PA Content Credentials, IPTC AI labels, and generator metadata. Runs in your browser and nothing is uploaded.
Drop an image here or click to browse
JPG, PNG, WebP, AVIF, plus MP4 and MP3. The file is inspected on your device and never uploaded.
checklistWhat gets checked
- C2PA Content Credentials: the cryptographic provenance record used by OpenAI, Adobe, Leica, Sony, and others.
- IPTC digital source type labels that formally declare AI generation.
- Embedded generator metadata from Stable Diffusion, ComfyUI, and common AI tools.
balanceHonest limits
A file with no marks is not proof of anything: social media and screenshots strip metadata. Google's SynthID watermark can only be read by Google's own tools, and no detector of any kind can identify every AI image.
What is the AI Image Checker?
An AI image checker inspects the provenance marks that AI tools and cameras now write into files. Since 2026 this ecosystem has real weight: OpenAI signs ChatGPT images with C2PA Content Credentials, Adobe Firefly and Google label their outputs, and cameras from Leica, Sony, Nikon, and Canon sign photos at capture. This checker reads those cryptographic records, the IPTC digital source type labels that formally declare AI generation, and the generation settings tools like Stable Diffusion embed in PNG files. Everything is read on your device, and the verdict is evidence based: it reports what the file itself declares instead of guessing from pixels.
How to use the AI Image Checker
- 1
Drop in a file
Images work best (JPG, PNG, WebP, AVIF), and MP4 video or MP3 audio with Content Credentials are read too.
- 2
Read the verdict
The result states one of four things: the file declares AI involvement, it carries signed provenance without AI marks, it holds weak AI hints, or it has no marks at all.
- 3
Open the evidence
When credentials exist you see who signed the file, when, with which tool, and what source type it declares. Metadata signals are listed with their strength.
- 4
Judge with the limits in mind
No marks proves nothing: sharing and screenshots strip metadata. Treat the result as evidence about this exact file, not a final ruling about the picture.
Frequently Asked Questions
Can this detect every AI generated image?
No, and no tool can. This checker reads the marks that generators and cameras write into files. When those marks were stripped or never written, it says so honestly instead of guessing. Pixel-guessing detectors exist, but their error rates are high enough that we chose not to present guesses as verdicts.
What are C2PA Content Credentials?
A cryptographically signed record of a file's origin, backed by Adobe, OpenAI, Google, Microsoft, and camera makers. As of 2026, ChatGPT images carry them, Google Search surfaces them, and cameras from Leica, Sony, Nikon, and Canon sign photos at the moment of capture.
The result says no marks were found. Is the image human made?
Unknown, and that is the honest answer. Most platforms strip metadata on upload, and screenshots produce entirely fresh files. Absence of marks is absence of evidence, not evidence of human origin.
Can it read Google's SynthID watermark?
No. SynthID is embedded in pixels rather than metadata and can only be decoded by Google's own tools, such as the Gemini app. That limit applies to every third party checker, so we state it instead of pretending otherwise.
Is my image uploaded for checking?
No. The C2PA toolkit runs as WebAssembly inside your browser and the metadata scanners read the file bytes locally. You can watch the Network tab while checking: the file never leaves your device.
Which AI tools leave marks this checker can find?
OpenAI ChatGPT and DALL-E images carry Content Credentials, Adobe Firefly signs its outputs, Google labels images through IPTC metadata, and Stable Diffusion and ComfyUI embed their generation settings in PNG files. Coverage grows as more tools adopt the standards.