FAUXLENS

AI Deepfake Detector

Check if a photo or video is real. Free to try, no account.

JPG PNG HEIC · MP4

Detects content from

  • OPENAI
  • MIDJOURNEY
  • Stability AISTABILITY
  • GEMINI
  • FIREFLY
  • FLUX.1
  • CANVA
  • IDEOGRAM
Private · Deleted the moment analysis ends

How Our Deepfake Detection Engine Works

Every scan runs up to three checks, in order, to decide whether an image or video was captured by a camera or made or altered by an AI model such as Midjourney, DALL-E, Stable Diffusion, Flux or Sora. The first decisive one settles it, and the report shows which one that was.

Content Credentials & Metadata

First, the file's own record. Some AI tools stamp a C2PA credential or a metadata label saying the image was generated, and that settles it. A full camera record can clear a photo. Missing metadata proves nothing, because screenshots and messaging apps strip it, so we never count its absence as evidence.

Invisible AI Watermarks

OpenAI and Google embed an invisible watermark in the pixels of images their AI tools make, designed to survive screenshots and compression. When we detect it, the report names which company's tool made the image. Tools that do not watermark are left to the next check.

AI Detection Model

Most images carry no credential and no watermark, so most answers come from the pixels. A specialist detection model scores the image and tells apart three outcomes: a real photo, part of a photo changed with AI, and an image generated from scratch.

Visual Review Fallback

If the detection model is unavailable, a multimodal vision model reviews the image instead: anatomy, hands and teeth, the direction of the light, text in the background. It writes out what it found, and that becomes the evidence line.

How the Verdict Is Reached

A credential or a watermark beats a pixel score, because it is a record rather than an estimate. Each check that found something gets its own line in the report. The confidence score is a weight of evidence, not a calibrated probability: 90% does not mean nine in ten such images are AI. Heavily compressed, forwarded or very small images are the weakest case, and the report says so when it sees one.

Why choose FauxLens?

Privacy-First Architecture

FauxLens deletes your upload once the check finishes and does not train on it. No account is needed for your first scan.

Covers All Major Models

Built for current generators, including Midjourney, OpenAI's image models, Gemini, Flux, Stable Diffusion, Sora and Veo. A brand-new model can be missed until detectors catch up.

Evidence With Every Verdict

Every verdict comes with a confidence score and the specific signals behind it. Use FauxLens as one step in a broader verification workflow.

Frequently asked questions

Our focus is AI generation. It flags edits made with AI tools. Ordinary editing without AI is not flagged, and it does not show where an edit is.
It shows how strongly the evidence points one way. It is not a calibrated probability: 90% does not mean nine in ten such images are AI. Read it together with the evidence lines.
FauxLens provides forensic indicators that can support an investigation. No AI detector should serve as the sole evidence in legal proceedings. Always pair results with expert human review.
It can weaken it. Re-compression and downscaling remove some of what the model looks for, and messaging apps strip metadata. The OpenAI and Google watermarks are designed to survive this. Upload the original file when you can.

Learn More About AI Detection

Is that photo or video real?

PrivacyTerms