Deepfake video check
Is this video a deepfake?
Upload a suspicious clip — a celebrity “investment” ad, a video message from a “relative”, or a recorded video call. GACS reviews 6 frames for face-swap and AI-generation signs and tells you what to do next. Try one check free without an account, or get more with a free GACS account.
Think of it as an AI video detector for still frames: it looks for signs a face was swapped or generated by AI, not for proof of who made the clip.
Still-frame review only — no voice, lip-sync or motion analysis. No-obvious-signs is not proof that the video or person is genuine.
What six samples can reveal — and miss
Visible clues: the reviewer can describe blurred face boundaries, mismatched lighting, warped background details or changes in glasses and facial features between the sampled images. Ordinary compression and editing can also produce unusual details.
Unseen intervals: the browser samples 6 points across the clip, rather than analysing every frame. A brief face swap or altered scene between those points may never reach the reviewer. The result applies to the samples, not every moment of the recording.
Missing evidence: the check cannot listen to a cloned voice, measure lip-sync, inspect continuous movement, establish who recorded the clip or verify a claimed endorsement. It does not authenticate a live call.
Before acting on a convincing clip
- Pause an unexpected payment or request for sensitive information, regardless of the detector score.
- Contact the person through a number you previously knew, or use an established workplace approval channel. Do not rely on another call arranged by the same unknown sender.
- Keep the original clip, source link and timestamps privately. Treat model observations as leads to investigate, not facts to publish about a person.
The FBI's December 2024 warning on AI-enabled fraud documents fake video and cloned audio and recommends independent identity checks. It is safety guidance, not an endorsement or accuracy assessment of GACS.
Common deepfake video scams
- Fake celebrity or news-anchor ads promoting an “investment platform”.
- Video calls from a “boss” or “relative” asking for an urgent payment.
- Romance-scam video messages using a stolen or generated face.
Read more: deepfake video-call scams · AI voice-cloning scams · deepfake photo check
Questions
How does the deepfake video check work?
Your browser takes 6 still frames spread across the clip and sends only those frames for review. The AI reviewer compares them for face-swap edges, details that change between frames, and AI-generation signs.
Does it check the voice or lip sync?
No. It reviews still frames only, not sound or motion. For suspected voice cloning, use the call-back and safe-word steps in our AI voice-cloning guide.
Is my video uploaded or stored?
The video file stays on your device. Only sampled frames and any context you enter are sent to an AI service for review. GACS does not save those frames as part of this check; usage is recorded to enforce your allowance. See the privacy policy for data handling.
Do I need an account?
You don't for your first check: visitors get one free photo or video check per day. Each video check reviews several frames, so it costs more to run than a link scan. Signing in lets GACS apply per-account allowances and reduce automated misuse; it does not prove someone is human. Free accounts get 2 checks per rolling 24 hours; Platinum, Talent Manager and Government Office plans get 10.
Is there a free AI video detector?
Yes. The GACS deepfake video check works as an AI video detector for still frames with one free check a day without an account, or 2 checks per rolling 24 hours with a free GACS account.
Can AI video detectors check voice clones?
This one cannot. It reviews still frames only and does not analyse audio. Treat urgent money requests in a familiar voice as unverified until you call back on a number you already have.
How accurate is it?
No independently validated accuracy rate is stated for this check. Its 0–100 score is a heuristic estimate, not a measured probability or forensic certification. A convincing fake can show no signs in the samples, and compression can create misleading artifacts.
Authoritative sources
Independent primary sources used to check and corroborate the guidance on this page.
- FBI Internet Crime Complaint Center (IC3)
Official US channel for reporting internet-enabled fraud and cybercrime.
- US Federal Trade Commission — ReportFraud
Consumer fraud reporting and published enforcement data.
- UK Report Fraud
UK national reporting centre for fraud and cybercrime.
- Canadian Anti-Fraud Centre
Canada's central repository for fraud reports and scam alerts.
