Cover the whole duration
Frames are spaced across the clip rather than taken only from the beginning.
Free Deepfake Video Detector
Review face swaps and AI-generated video using timestamped frame evidence.
Up to 100 MB, no signup, uploads are not retained.
Video · Drop files or select multiple to begin bulk analysis
Video
File size limit: 100MB (MP4, MOV)
Video · Drop files or select multiple to begin bulk analysis
Video
File size limit: 100MB (MP4, MOV)
Uploads Not RetainedCurrent request only
Up to 10 FramesAcross the full duration
Up to 100MBMP4, MOV, WebM
Actual analysis path
What the video detector actually reviews
The browser samples up to 10 frames across the full clip. Each frame keeps its timestamp so repeated anomalies can be compared across time.
Look for repeated conflicts
Face boundaries, objects, lighting and motion are reviewed at multiple timestamps.
Keep single-frame noise uncertain
One compression artifact is not enough for a high-risk result; repeated evidence matters more.
StartFull clip timelineEnd
F1
F2
F3
F4
F5
F6
F7
F8
F9
F10
The percentages illustrate the current spacing rule. This is a method explanation, not a fabricated result from a specific upload.
Video Deepfake Detection — Is This Video Real or AI?
DeepFakeCheck helps you review face-swap, AI-generated and AI-edited video. Upload an MP4, MOV or WebM file up to 100 MB. The browser samples up to 10 timestamped frames across the full duration, then the analysis model reviews repeated visual inconsistencies. No signup is required, usage limits apply, and uploads are not retained.
How deepfake video detection works
The current pipeline is deliberately simple and transparent:
- Uniform frame sampling. Up to 10 frames are spaced across the full clip, not just the opening seconds.
- Timestamped review. Each sampled frame keeps its original timestamp so a visible issue can be checked across multiple moments.
- Visual reasoning. The model reviews face boundaries, object continuity, lighting and physically implausible changes across the sampled frames.
- Conservative conclusion. A single compression block or blurred frame is not enough for a high-risk result. Repeated, independent conflicts carry more weight.
What the review may surface
The sampled evidence may surface patterns associated with several common manipulation types:
- Face swaps: repeated boundary, skin-tone or lighting conflicts around the face and hairline.
- Fully generated video: objects, anatomy, text, reflections or shadows that change implausibly between samples.
- AI editing: a localized region that changes differently from the surrounding scene.
Common use cases for the video detector
- Newsrooms verifying viral political clips before broadcasting.
- KYC and fraud teams spotting deepfake video during identity verification — a fast-growing attack vector in fintech.
- Influencer managers and brands monitoring whether the talent they represent is being deepfaked in scam videos.
- Family members checking suspicious video calls — voice-clone plus face-swap scams targeting older relatives are now common.
Video detection FAQ
How much of the clip is reviewed? The current browser pipeline selects up to 10 frames across the entire duration. Because it does not inspect every frame, a very short manipulation between samples can be missed.
Does it work on heavily compressed social video? It can still review visible inconsistencies, but compression and resizing remove forensic detail and may also create false artifacts. Treat these results with lower confidence.
Can it prove a video is authentic? No. A low-risk result only means the sampled frames did not contain strong evidence. It does not certify every frame or the original source.
What does "high risk" mean? It means the sampled evidence contains multiple patterns associated with AI generation or manipulation. It does not establish malicious intent.
Is my video private? The video is transmitted over HTTPS for the current analysis and is not retained by DeepFakeCheck. We never train models on uploads.