Someone sent you a video and something feels off. Our AI video detector analyzes a clip frame by frame, checking motion patterns, facial consistency, and generative artifacts, then returns an authenticity verdict with a confidence score in seconds. Not a guess.
MP4, MOV, or WEBM. Files deleted within 60 seconds of analysis.
No technical background needed. Upload a clip, wait a few seconds, read your result.
Drag and drop or browse. MP4, MOV, and WEBM are supported. Your video is used for analysis only and deleted from primary storage within 60 seconds of analysis, unless you opt into retention.
The detector inspects each frame for the fingerprints generative models leave, then compares frames against each other. Real footage moves in physically consistent ways; AI video drifts. That temporal check is where many fakes fall apart.
A clear verdict (Authentic, Likely Synthetic, or Inconclusive) plus a TrustScore from 0 to 100. A 95 is strong evidence; a 60 means the clip deserves more scrutiny. We show the score because honesty beats false certainty.
In one widely reported case, a finance employee at engineering firm Arup wired $25.6 million after a video call where every other participant was a deepfake, as reported by CNN. Our detector covers the major families of fake video in circulation.
| Generator family | Examples | What we look for |
|---|---|---|
| Text-to-video models | Sora, Runway, Pika, Kling | Motion artifacts, physics errors, texture drift between frames |
| Face-swap deepfakes | GAN and diffusion face replacement | Blending boundaries, mismatched lighting, identity inconsistency |
| Lip-sync manipulation | Dubbed or re-voiced clips | Mouth shapes that do not match the audio, jaw and teeth artifacts |
| Synthetic avatars | AI presenter and talking-head videos | Rigid head motion, repetitive gestures, uncanny eye behavior |
What we cannot promise. Detection is probabilistic, not magic. Heavily compressed, cropped, or re-encoded clips lose the detail our models rely on, which lowers confidence. Very short clips give the temporal checks less to work with, and brand-new generators can evade detection for a window until models are updated. Any tool that claims it catches everything is overselling.
Our models reach high accuracy across video, image, and audio testing: across our evaluation sets, the overwhelming majority of files are classified correctly. That does not mean every verdict is certain, which is why each result ships with a confidence score.
Independent research backs that caution. The DeepFake Detection Challenge dataset paper found detectors that performed well on training data scored markedly lower on unseen, real-world fakes. That is exactly why we update our models as new generators appear and publish honest numbers instead of chasing headline claims.
| What you get | DeepfakeDetector.ai |
|---|---|
| Detection accuracy | high accuracy |
| Formats covered | Video, image, and audio in one account |
| Free tier | 50 detections per month, videos up to 2 minutes per detection |
| API access | Yes, on every paid plan from $49 per month |
| Browser tool | Chrome extension |
| File handling | Deleted from primary storage within 60 seconds of analysis |
Start checking videos without paying anything. A free account includes 50 detections per month across video, image, and audio, with videos up to 2 minutes per detection. For most people verifying the occasional clip, the free tier is all you need.
| Plan | Price | Detections | Clip length |
|---|---|---|---|
| Free | $0/mo | 50/mo | Up to 2 min |
| Starter | $49/mo | 1,000/mo | Up to 10 min |
| Business | $199/mo | 5,000/mo | Up to 10 min |
| Enterprise | $599/mo | 20,000/mo | Up to 10 min |
Every paid plan includes API access, PDF and CSV export, and the same models through our deepfake detection API. Checking a photo instead? Try the AI image detector.
A scan is the fastest route to an answer, but your own eyes catch plenty. One sign alone proves nothing; several together are a strong signal.
Unnatural blinking (too rare, too regular, or never quite closing) and mouth or teeth glitches where lips fall out of sync with the audio.
Flickering or smearing around the hairline, jaw, and ears on head turns, and face shadows pointing a different direction than the room.
Straight lines that bend, objects that melt as the camera moves, extra fingers, and morphing rings or glasses between frames.
For the full walkthrough, read how to tell if a video is AI generated, or start with the basics in how to detect a deepfake.
Everything people ask before their first scan. If yours isn't here, mail hello@deepfakedetector.ai.
Combine manual checks with a detector scan. Look for unnatural blinking, lip-sync errors, and warped backgrounds, then upload the clip for a frame-by-frame verdict. Our guide on how to tell if a video is AI generated walks through every sign.
Yes. A free account includes 50 detections per month, covering video, image, and audio, with videos up to 2 minutes per detection. No card required.
Yes. Our models detect output from current text-to-video generators, including the Sora, Runway, Pika, and Kling families. Verdicts are probabilistic, so a confidence score accompanies every result, and coverage is updated as new generators appear.
MP4, MOV, and WEBM files are supported, with clips up to 10 minutes long per detection on paid plans and up to 2 minutes on the Free plan.
No. Files are deleted from primary storage within 60 seconds of analysis completion, unless you opt into retention. A SOC 2 audit is in progress as part of our security program.
The next suspicious clip you receive does not have to stay a mystery. Run it through the AI video detector and get a verdict with a confidence score in seconds. 50 checks every month, no card required.