How to use this tool
Checks the image with two neural models and estimates how much it resembles AI-generated imagery.
- 1Choose one JPG, PNG or WebP image.
- 2Wait for the check. Browser processing downloads the models on its first run.
- 3Read the result, then check the limits listed below before deciding what it means.
Best for
A first-pass review of an image when you want an additional pixel-based signal before human or provenance review.
Privacy
When server processing is available, your image is sent to our detector and processed in memory without being saved. Otherwise, the check runs in your browser. Image contents are never included in analytics. Read our privacy policy
What the neural detector does
Detectra and Frontier models check the image pixels and produce an estimated AI-generated likelihood. The result does not rely on the filename, EXIF tags or C2PA credentials.
Read the result as a signal, not proof
High and low scores describe which class the model found more consistent with the pixels. The middle band is deliberately inconclusive. A detector can miss new generators, flag edited photographs or artwork, and change its result after cropping, resizing or compression.
Inspect the metadata separatelyRemove embedded metadata from a copy
Model, browser runtime and privacy
Both models are pinned to specific revisions. When our server is available, your image is uploaded and processed in memory without being saved. Otherwise, the models run in your browser and are cached there for later checks. The result shows the runtime used. Image contents are never sent to analytics.
When a detector is not enough
For a consequential decision, compare the original file, verified provenance, source context and human review. Missing C2PA credentials are not evidence that an image is synthetic, and this tool does not make an authorship or authenticity determination.
Limitations
- The model can be wrong. Cropping, resizing, compression, editing, screenshots, human artwork and newer generators can change the result.
- An inconclusive result means the model found weak evidence; it does not prove that an image is real or AI-generated.
- This page does not verify C2PA credentials, identify a generator, or certify authenticity. Use provenance and human review separately for consequential decisions.
- Browser processing downloads both models and caches them. Speed depends on your device, memory and browser support.
Questions
Is this an AI image detector or a metadata checker?
It checks image pixels with Detectra and Frontier neural models. Metadata is not used to decide the result.
Is the image uploaded?
When server processing is available, yes. The detector processes your image in memory without saving it. If the server is unavailable, processing runs in your browser. Image contents are not sent to analytics.
What does the percentage mean?
It is the model's estimated likelihood that the pixels resemble AI-generated imagery. It is not a measured probability of authorship and is not proof.
Why is the result inconclusive?
The score is in the middle band where the model's evidence is weak. Editing, compression, screenshots, unusual artwork and unfamiliar generators can all make classification uncertain.
Which model does the detector use?
It combines Detectra v3 and Frontier models pinned to specific revisions. The result shows whether processing used our server or your browser.
Can I use this as evidence?
No. Treat it as one review signal only. For important decisions, compare the original file, verified provenance, source context and human assessment.
Related tools
- EXIF Viewer
View the EXIF, GPS, camera and editing details stored inside a photo. Read-only and processed in your browser.
- Remove Metadata
Remove GPS, camera details and hidden photo information. Keep the image quality.
- Redact Image
Cover private text, faces or numbers with solid black boxes.
- Image Editor
Edit, crop, filter, draw on, and add text to JPG, PNG, and WebP images.