Upload
Choose a JPG, PNG, or WebP file. Your browser reduces the longest edge to 2048 pixels before the image is sent, which keeps the request fast and below the server limit.
Evidence first · No signup
AI Photo Detector
Upload a photo to check whether it is more likely AI-generated, AI-edited, or camera-origin. Get a confidence score, provenance details, and plain-language reasons in seconds. Free, no account, no credit card.
JPG · PNG · WebP5 free scans/dayNo image history
Drop a photo here
JPG, PNG, or WebP. Five free scans per browser each day.
3-step
Upload, analyze, read the evidence.
2048px
Large mobile photos are resized before upload.
2 layers
Metadata checks run first, then a visual model when needed.
5/day
The usable free allowance is enforced per browser.
How it works
Three steps from upload to evidence
Analyze
FauxGuard reads EXIF and provenance markers first. If those signals cannot answer the question, the image is queued for a visual generative and deepfake check.
Read the result
You get a probability, confidence band, generator family when available, provenance status, and at least three reasons behind the score.
Signals we inspect
More useful than a black-box percentage
A detector is only helpful when it shows its evidence. FauxGuard separates local provenance from model inference so you can see what was actually found.
Metadata provenance
We inspect EXIF camera fields, editing software tags, C2PA Content Credentials, and known AI watermark markers before sending an image to a model.
Generator fingerprints
Known generator names and signatures can survive in file metadata. When one is found, the result identifies the likely model family instead of returning only a score.
Visual model signals
A third-party visual detection model evaluates generative and deepfake signals when local metadata cannot settle the question. The result is normalized into one plain-language score.
Editing and tamper cues
Photoshop, Lightroom, GIMP, and similar software tags help distinguish an AI-generated image from a real photo that was manually edited.
Generator coverage
Generators FauxGuard can identify from metadata
This list is data-driven and can be expanded without rewriting the homepage. Metadata is not present in every image, so a missing match reduces confidence rather than proving the image is real.
OpenAI image models
Midjourney
Stable Diffusion ecosystem
Adobe Firefly and Photoshop generative tools
Google Imagen and Gemini
Flux and other modern diffusion models
Ideogram, Leonardo, and Canva AI
Real-world use
Where an AI image signal helps
Fake social media identities
A convincing profile photo can make a new or automated account look established. Attackers use generated portraits because they are fast to produce, easy to customize, and do not require a real person's consent. That creates risks for dating platforms, community moderators, recruiters, and anyone deciding whether to trust a stranger online. An AI photo detector can add a useful signal when the profile has little history, but the result must not be treated as proof. A real-looking face may belong to a real person, while a generated image may have no detectable metadata. FauxGuard shows whether camera provenance or a known generator signature was found, then explains what remains uncertain. Review the account's behavior, links, posting history, and other context alongside the image result.
Document and identity fraud
Edited identity photos, altered statements, and synthetic supporting images can be used to pass a casual review or pressure a real person into acting. In a fraud workflow, the attacker usually needs an image to look plausible for only a few minutes. Compression from messaging apps can remove file metadata, so a detector should not pretend that a missing camera tag proves manipulation. FauxGuard separates provenance checks from visual model signals and avoids asserting that an image is fraudulent. The safest process combines image analysis with document verification, liveness or video checks, source validation, and a human review path. Never reject a legitimate customer or applicant solely because an automated score is high, and never approve a high-value request solely because a score is low.
Fake news and misleading captions
A synthetic scene can be paired with a real event, an old photo can be relabeled, or a genuine image can be edited before it spreads. Visual detectors are most useful when they are part of a wider verification workflow: search for earlier copies, inspect the publisher, compare time and location, and ask whether the original file is available. FauxGuard reports a confidence band and the evidence behind it rather than issuing a final verdict. That distinction matters because metadata can be stripped by a platform, altered by an editor, or preserved from a real camera even after the content has changed. The goal is to help a reader slow down and ask better questions, not to automate a public accusation or replace editorial judgment.
Marketplace and ecommerce returns
Sellers and support teams increasingly receive photos as proof in disputes over damaged products, missing items, or returns. Generated or heavily edited evidence can make a claim harder to resolve, while image processing on a phone can make authentic photos look inconsistent. An AI image detector can flag a likely generative signal, identify a known generator family when metadata survives, and show whether editing software was named in the file. It cannot establish what happened to a package or prove intent. Teams should keep the original upload where possible, preserve the request record, compare the image with order details, and give the customer a clear appeal path. The detection result is one review signal inside a documented process, not a substitute for policy or customer service.
FAQ
Questions before you upload
How accurate is an AI photo detector?
Accuracy depends on the image, the generator, compression, and whether provenance metadata is present. FauxGuard separates high, moderate, and low confidence so you can judge how much weight to give a result. No detector can prove origin with perfect certainty, and an image with no visible AI signal is not automatically authentic.
Does FauxGuard work on edited photos?
Yes. FauxGuard checks for common editing software tags and can return a manual-edit signal even when the base image came from a camera. That is different from saying the whole image was generated by AI, so the result explains the distinction rather than hiding it behind a single label.
Can the result prove that an image is fake?
No. A detection result is evidence, not a legal finding. Metadata can be removed or forged, compression can erase clues, and a real photo can look unusual after editing. Use the score, reasons, provenance, and context together before making a decision.
What image formats are supported?
The detector accepts JPG, PNG, and WebP images. Your browser resizes large images to a maximum edge of 2048 pixels and exports WebP before upload when the browser supports it, which keeps mobile uploads fast.
Is the image stored or used for training?
FauxGuard does not create an account history or save your upload in its own database. The image is processed for the current detection request. If the visual model is required, the processed image is sent to a third-party visual detection service; server credentials stay on the FauxGuard side.
Do I need an account or a credit card?
No account is required for the free detector. The free allowance is five scans per browser per day and resets each day.
Why does a result sometimes say model detection is unavailable?
The metadata check runs locally and can still return provenance or software clues. If the upstream model is unavailable, rate-limited, or missing credentials, FauxGuard returns that partial evidence with a clear notice instead of failing the whole page.
Ready to check an image?
Start with the free detector. Compare paid API capacity on the pricing page when you are ready to test a workflow.