How to detect a deepfake video call in real time
Live face-swap and lip-sync tools now run on consumer laptops. Here’s the 60-second checklist for verifying a suspicious video call — plus the visual and audio signals our Trust Score uses to flag real-time deepfakes.
Why video calls are the new attack surface
Real-time deepfakes moved from research demos to off-the-shelf tools in 2024. Scammers now join Zoom, Teams, and WhatsApp video calls wearing a stolen face — usually a CFO, an IT admin, or a family member — and ask for a wire, a password reset, or an urgent favour.
The good news: live deepfakes are still fragile. Side-profile views, hand occlusion, and quick lighting changes all break the model in ways your eyes can spot in seconds — if you know what to ask for.
The 7-step real-time checklist
Run this list during the call, before approving any payment, credential change, or sensitive action.
- 1
Pause the call before any decision. Real urgency survives 60 seconds of verification.
- 2
Ask them to turn their head fully sideways — deepfake models break at extreme profile angles.
- 3
Ask them to wave a hand slowly in front of their face — occlusion exposes the swap.
- 4
Ask them to hold up a piece of paper with a number you just said.
- 5
Watch for lip-sync drift, especially on plosive sounds (p, b, t).
- 6
Hang up and call them back on a number you already have saved.
- 7
Use a shared safe-word or a question only the real person could answer.
What our deepfake video detector checks
Behind the Trust Score, four families of signal do the heavy lifting on recorded video clips.
Real eyes saccade — they make tiny, constant jumps. Deepfake eyes often drift smoothly, blink at the wrong cadence, or fail to track moving objects on screen.
Generated skin tends to be too smooth or shows repeating micro-patterns under bright light. Hair edges, jawlines, and ears are where most live deepfakes fall apart first.
Real-time mouth synthesis usually lags audio by 80–200 ms, and consonants like 'p' and 'b' don't close cleanly. A 5-second clip is enough for a detector to measure the drift.
When the person moves, shadows on the synthetic face don't follow the room's light source. Ask them to wave their phone screen across their face — real shadows track, deepfakes don't.
Set a verification protocol with your team
The strongest defence against video deepfakes is procedural, not visual. Agree the rules before you ever need them.
- No money moves on the strength of a video call alone — confirm via a second channel.
- Agree a team safe-word for any “urgent” finance request.
- Train everyone to ask the profile-turn and hand-wave tests without embarrassment.
- Record suspicious calls (where legal) and run them through a Trust Score check.
Frequently asked
Can deepfake video calls really happen live?
Yes. Real-time face-swap and lip-sync tools now run on consumer GPUs. A scammer can join a Zoom or WhatsApp video call and impersonate a colleague or family member convincingly for several minutes.
How can I tell if a video call is a deepfake in real time?
Ask the person to turn their head fully sideways, wave a hand in front of their face, or hold up a written note. Real-time deepfake models still struggle with profile views, occlusion, and unexpected hand motion.
Does the Trust Score work on video calls?
Yes. Record a short clip (where legal) and upload it — our deepfake video detector returns a Trust Score plus the specific signals it flagged, like lip-sync drift or texture inconsistencies.
What if the caller refuses to do the verification gesture?
Treat that as a red flag. A real colleague or family member will humor a quick sanity check, especially when money, credentials, or access is on the line.
Got a video clip you’re unsure about?
Upload it — we’ll check for deepfake signals and return a Trust Score in under a minute. Free, anonymous, no account needed.
Check a video clip now