Start with the original image file and inspect its metadata for declared AI tools, saved generation parameters or Content Credentials. Then investigate the source and context. No single missing tag, visual oddity or detector score can establish whether an image is real.
AI Image Checker
Inspect image metadata for AI-generation clues, saved workflows and credential markers. Metadata evidence only: no pixel-model predictions or AI probability scores.
Static JPG, PNG or WebP · up to 20 MB, 8 megapixels and 8192 px per edge. Read-only checks; no image uploads.
Choose an image to begin.
Metadata can be edited, copied or stripped. These clues cannot prove whether an image is real or AI-generated.
Evidence found
Content Credentials markers
Finding a marker is not cryptographic validation. This checker does not verify signatures, signers, provenance claims or external manifests.
Image information
Recorded image density
DPI is commonly used in searches; image density is measured in pixels per inch (PPI). A density tag does not tell you whether an image is sharp.
Calculate print size
Inspection notes
This checks common EXIF fields, PNG text, supported XMP fields and selected credential markers. Unsupported or removed metadata cannot be recovered. It is not a complete file-integrity, copyright or authenticity check.
View inspected metadata
Metadata may include private names, dates or prompts. The report includes these fields.
What this AI image checker does
This version reads metadata locally. It looks for recognised generator tags, structured generation parameters, saved ComfyUI workflows and selected credential markers. It does not run a pixel-detection model, provide an AI probability or validate credential signatures.
Use the original file
Ask for the original download or export when available. Screenshots, social-media copies, compression and re-saving can remove metadata. A file without useful metadata may be a camera photo, an AI image or an edited combination.
Interpret the result carefully
- AI-related metadata found: inspect the named field. It may describe a tool or workflow, but metadata can be copied or changed.
- Unable to determine: no recognised clues were found in the inspected fields. This is not a verdict that the image is human-made.
- Credential marker found: a possible provenance container or reference was detected. This tool has not validated its signature, signer or claims.
Metadata evidence versus model predictions
Metadata inspection reports information embedded in the file. A pixel-based detector instead estimates origin from learned visual patterns. Those are different kinds of information and should be labelled separately. This checker’s model result is always Not run; it does not turn metadata matches into percentages.
What are Content Credentials?
Content Credentials can record information about creation and editing history. Proper validation checks signatures and the relationship between the record and the file. Their presence does not establish whether the depicted event really happened, and their absence does not imply a fake. This checker identifies selected markers only.
What else can I check?
Look for the earliest known source, the creator’s explanation and corroborating evidence. Compare alternate versions for differences. Visual inconsistencies can suggest questions to investigate, but they are not reliable proof on their own. Our tools do not perform a reverse-image search or contact external services with your image.
Can AI detectors be wrong?
Yes. A model’s performance can depend on its training data, the generator used and later editing. Treat a detector result as an estimate with limitations, not a guarantee. Avoid presenting an untested numeric score as a probability.
Save the evidence, not a verdict
The downloadable JSON report records the inspected fields and limitations. It may contain private names or generation prompts, so review it before sharing. Your image stays unchanged and is not uploaded.
Read more: Content Authenticity Initiative FAQs and research on AI image detector generalisation.
Behind the result
This is a rule-based inspection of embedded metadata, separate from any model that predicts origin from pixels.
Read supported metadata
Bounded parsing examines common EXIF text, PNG text including supported compressed fields, and selected XMP properties. Unsupported or excessive metadata is reported as a limitation.
Identify specific clues
The checker looks for recognised AI-associated software tags, structured generation parameters, ComfyUI workflow patterns and AI digital-source declarations. Ordinary mentions in image descriptions are not treated as generator evidence.
Report credential markers
Selected C2PA-related markers are reported as present but not validated. Signatures, signer trust, manifest claims and external provenance are not checked.
Keep uncertainty visible
No pixel detection model runs. No score is calculated. Missing clues produce Unable to determine, while detected metadata is labelled as evidence that can be changed or copied.
Explore the details: How Content Credentials workC2PA specifications