Knowledge base

AI Image Detection FAQ

Detailed answers about FauxGuard, image metadata, AI-generation scores, editing detection, deepfake checks, privacy, and the limits of automated verification. Start here if you want context before acting on a result.

Using FauxGuard

What is FauxGuard?

FauxGuard is an image inspection tool that estimates whether a still image is more likely AI-generated, AI-edited, or camera-origin. It returns a probability, a confidence band, provenance details, and reasons for the result. It is designed for first-pass review and education, not for making a legal or factual determination about a person.

What image formats can I upload?

FauxGuard accepts JPG, PNG, and WebP images. Large images can be reduced by the browser before upload, which keeps the check faster on mobile connections. If a platform has converted your image into a screenshot or a heavily compressed social-media file, some metadata may be missing before it ever reaches FauxGuard.

Do I need an account?

No account is required. Buy 20 image checks for $4.99 as a one-time purchase, and save your recovery code to restore your balance on another device. There is no daily free allowance, subscription, personal dashboard, or saved scan history.

Where is the detector itself?

The upload area is on the FauxGuard home page. Focused pages such as Photoshop Detection, Deepfake Detector, and Is This Image AI Generated? explain how to interpret a result for a particular question, then link back to the same detector.

Understanding the result

What does the AI-generation probability mean?

The probability is a 0–100 estimate produced from the available provenance and visual-model evidence. It is not a percentage chance that a particular person used a particular tool. A high score means the available signals point toward generative content. A low score means the available signals were weak or pointed toward a camera workflow.

What is the difference between a score and confidence?

The score describes the estimated direction of the evidence. Confidence describes how strong or consistent the evidence is. A high score with high confidence is different from a high score with low confidence, and a 50/100 score can be more informative than a dramatic number when the file contains conflicting signals.

What does “likely AI-generated” mean?

It means the probability and confidence are high enough to support a likely conclusion, while still leaving room for error. FauxGuard avoids saying “this image is AI-generated” because metadata can be altered, models can be wrong, and a single file does not prove who created it or why.

Why does the result include reasons?

A score without an explanation is difficult to review. The reasons show whether the result came from metadata, a known generator signature, a visual model, or a combination of evidence. That helps a person decide whether to request an original file, compare another copy, or conduct a manual review.

Metadata, editing, and deepfakes

What metadata does FauxGuard inspect?

FauxGuard reads camera make and model fields, editing software tags, Content Credentials, SynthID-related markers, and known generator signatures when those fields survive in the uploaded file. Metadata is helpful but not permanent. Messaging apps, screenshots, exports, and social platforms can remove it.

Can FauxGuard detect a manually edited photo?

It can identify editing software names and other provenance clues in some files, and a visual model can add evidence when metadata is inconclusive. A Photoshop tag shows that an editor was used, not that the image is false. Cropping, color correction, retouching, and compositing are legitimate workflows.

Can FauxGuard detect a deepfake?

The current service checks still images for generative and deepfake-related visual signals. It cannot analyze video, audio, or a live stream. A single frame from a video can be checked as an image, but the result applies to that frame rather than the whole recording.

Can a detector identify the generator or model?

Sometimes. A readable signature, Content Credential, or software tag may identify a model family. A visual model can indicate that generative patterns are present without naming the exact tool. If the file has been re-encoded or stripped, the generator may be impossible to identify.

Privacy and responsible use

Does FauxGuard store my image?

FauxGuard does not provide a user account history or an image gallery. The image is processed for the current request and may be sent to a third-party visual detection service when visual analysis is needed. Do not upload sensitive material unless you are authorized to share it with the services described in the Privacy Policy.

Can I use a result as proof in a dispute?

No. The result is informational and may be wrong. A legal, employment, insurance, marketplace, or reputational decision should include source evidence, human review, a documented policy, and an opportunity for the affected person to respond.

What should I do when the result is uncertain?

Treat uncertainty as useful information. Preserve the original file, look for an earlier copy, check the publisher and caption, compare independent sources, and ask for an original camera file when the decision matters. A partial or low-confidence result should lead to more review rather than a forced answer.

Need to check a specific image?

Upload a JPG, PNG, or WebP file and compare the score with the provenance and reasons. For a more focused answer, use the Photoshop, deepfake, or question-based pages.

Open the detector