Deepfake detector for audio, video and image.
Upload a file. Get a synthetic-media verdict in under a minute, powered by the same detection stack Diopter ships to enterprise security teams.
Know what is real or fake instantly.
Every result delivers a verdict, a confidence score, and the signals that prove it.
- A verdict: Likely Synthetic or Real.
- A detection score and confidence level.
- An overview of suspect regions detected in the file.
- Signal-by-signal breakdown
- Analyzer timeline
- Unlisted shareable link for colleagues
- Request expert review
Three steps from file to verdict.
Agree and upload
Drag in voice notes, video clips, or images. Files up to 200 MB are encrypted in transit.
Multi-signal analysis
We check your file for deepfake signatures on temporal, spectral, generative, and auditory levels at once.
- Temporal
- Blink rates, micro-expressions, and motion artifacts
- Spectral
- Formant irregularities below human perception
- Generators
- Fingerprints matched against HeyGen, ElevenLabs, Stable Diffusion, Sora, and more
Get your verdict
A verdict, a detection score, a confidence level, and a suspect-region overview. Sign up for the signal-by-signal breakdown and expert review.
Who uses a deepfake detector.
Five situations where a single check changes the decision that follows.
Scam detection
A voice note, video clip, or image arrives from a trusted source with an urgent ask. Voice-clone and deepfake-video scams have cleared wire transfers and bypassed identity checks. Run the file before you act, because an executed transfer may not be recoverable.
Executive & vendor authorization
A recorded call or voice note from an executive pushes a payment or exception outside normal channels. The $25.6M Hong Kong incident in 2024 ran exactly this play. Treat any recorded authorization like a wire transfer: verify it is real before you execute.
Hiring fraud detection
A remote candidate interviews over video for a role with system access, financial operations, or sensitive data. State-sponsored rings are placing synthetic candidates on payrolls, and deepfake interview fraud rose 220% year over year in 2024. Check the recording before you extend an offer.
Identity verification
A liveness-check video, onboarding selfie, or ID scan clears your biometric match on paper. AI-generated faces now pass standard liveness checks often enough that major KYC providers added secondary layers. Run submitted media as a second signal before the identity is accepted.
Social media & media integrity
A video or image of a public figure, colleague, or contact is circulating and something is off. Deepfakes spread faster than corrections. Check the file before you share, report, or respond.
Why deepfake detection matters.
The incidents are public. The losses are documented.
In 2024, a finance worker at a Hong Kong multinational joined a video call where every face on the screen was fake. The CFO. The colleagues. The counterparties. All of them deepfakes. He approved 15 wire transfers worth $25.6 million before anyone realized the only real person on the call was him.
In 2020, a bank manager wired $35 million on the strength of a voice. It belonged to a director he had spoken to before, and the voice sounded right. Forensics confirmed the call was synthetic after the money was already gone.
Eleven seconds of audio is enough. That is all current voice-cloning tools need to rebuild a voice that fools a human listener more than 85% of the time. It does not sound close. It sounds like them. That is the point.
Check it before you act on it.
Trained against the tools attackers actually use.
Retrained against updated outputs before a new generator reaches widespread deployment.
The deepfake and voice-synthesis tools most commonly deployed in fraud each produce a generation signature. Diopter's model is retrained against updated outputs before any new tool reaches widespread deployment.
Files up to 200 MB on the public detector.
Deepfake detector FAQ.
Deepfake detection tells you whether a video, image, or audio file was generated or altered by AI. You get a verdict, Likely Synthetic or Real, a detection score, a confidence level, and an overview of suspect regions.
All three media types. Video: face swaps and synthetic faces. Images: AI-generated and manipulated stills from diffusion and GAN models. Audio: cloned voices and synthetic speech from ElevenLabs, HeyGen, XTTS, RVC, and their equivalents.
The detector reads several signal layers at once. In video: facial geometry and lip-sync. In audio: formants, pitch, and spectral patterns against known synthetic signatures. In images: pixel-level artifacts and generator fingerprints. One file, multiple verdicts cross-checked.
MP3, WAV, and M4A for voice, MP4, MOV, and WEBM for video, and JPG, PNG, and WEBP for images. Files up to 200 MB are accepted on the public detector.
No. Your first detection is free and returns a verdict, confidence score and suspect-region overview with no sign-up. Create a free account for the signal-by-signal breakdown and expert review.
Yes. Fraud, KYC, and compliance teams verify suspicious media before acting. The enterprise platform fits the detection into your triage, case management, and identity workflows.
That is exactly the threat it was built for. Cloned executive calls, deepfake authorization requests, synthetic job candidates, AI-voice vendor fraud: each produces a file you can submit. Check the recording before you act on the instruction it contains.
The public detector scans recorded media. For real-time protection on Zoom, Teams, Google Meet, and phone calls, talk to our team about the Diopter enterprise platform.
Yes. The full platform adds live-call monitoring, conversation scoring, batch processing, and team account management. Book a 30-minute walkthrough.
Real-time deepfake protection for Zoom, Teams, and phone calls.
The detector you just used is part of the Diopter platform, built for live-call voice and video defense at the executive, finance, and hiring layer.