Can you tell if a video call is a deepfake? Sometimes, but not reliably. Real-time face-swapping tools can now put a familiar face onto a stranger during a live call, and the visual glitches that used to give them away are getting harder to see. The check that still works is procedural: end the call, contact the person through a channel you already trust, and never move money or share codes during the call that asked for it.
This guide covers the visual signs worth knowing, why they are weak evidence on their own, and a short routine you can teach your parents or your accounts team in five minutes.
How is a live deepfake video call put together?
A live deepfake call has three ingredients. The first is source material: photos and video clips of the person being impersonated, usually taken from public social media. The second is a face-swap or face-reenactment model that maps that face onto whoever is sitting in front of the scammer's camera, frame by frame. The third is a virtual camera, which feeds the altered picture into an ordinary video app as if it came from a normal webcam. Nothing about the app itself is hacked; the app simply receives a fake picture.
Voice is often handled separately. Some scammers use a cloned voice, others stay silent and type, and many simply blame a bad connection. In the scams widely reported in Vietnam, a common pattern was a hijacked Facebook or Zalo account messaging relatives to borrow money, followed by a few seconds of blurry video to “prove” it was really them. The short, low-quality call was the point: just enough to remove doubt, not long enough to be examined.
Quality depends heavily on conditions. Face-swap models are trained mostly on front-facing, well-lit faces. They struggle more with side profiles, fast movement, objects passing in front of the face, unusual lighting and high resolution. That is why scammers prefer dim rooms, shaky cameras and calls that end quickly.
The important point for defenders is that the effort required has dropped. What once needed a studio and hours of processing can now run on a decent gaming laptop. You do not need to be famous to be impersonated; a few dozen public clips of your face may be enough to make a convincing few seconds.
None of this means every glitchy call is fake. Real calls freeze, blur and lag all the time on mobile networks. Visual oddities are a reason to verify, not a verdict.
Visual and audio signs of a deepfake call, and their limits
Edges around the face. Look where the face meets hair, ears and neck. Blurring, a shimmering outline, or skin tone that changes at the jawline can be signs of a swap. Limit: video compression on a weak connection produces similar smearing.
A hand passing in front of the face. Asking the person to wave a hand slowly across their face, or touch their nose, used to break many models, which would flicker or reveal the real face for a moment. Newer tools handle this better, so a clean result proves little.
Turning to a full side profile. Ask them to turn their head sideways and hold it there. Profiles are harder to fake because training data is mostly frontal. Watch for the face warping, flattening or lagging behind the movement of the head.
Lips and voice out of step. Speech that does not match mouth shapes, especially on sounds like b, m and p where the lips must close, is worth noting. Network lag causes the same thing, so compare with the rest of the call.
Eyes, glasses and teeth. Blinking that looks mechanical, reflections in glasses that do not match the room, or teeth that blur into a single block are classic artefacts. They are also the first things tool makers fix.
Lighting that does not move. If the person walks near a window or a lamp and the light on their face stays oddly constant, the face may be generated separately from the scene around it.
A story that resists checking. Phone broken, camera bad, battery low, cannot talk long, in a hurry, do not tell anyone. These are not visual clues, but they are the most consistent sign across scams, because the scammer needs the call to stay short.
Pressure to act inside the call. Any request to transfer money, read out an OTP, install an app or share your screen during that same call is the real red flag. A genuine relative or boss can wait ten minutes for you to call back.
A five-minute verification routine that actually works
Visual checks can be beaten; procedures are much harder to beat. Agree on this routine with family and colleagues before you need it.
- Say you will call back. Use a neutral line such as “I will call you right back on your usual number.” A real person rarely objects. A scammer usually pushes back or invents a reason why that number will not work.
- Hang up and use a number you already had. Call the phone number saved in your contacts before this conversation, not one sent in the chat. If the social media account was hijacked, the attacker controls that account, not the person’s SIM card.
- Ask something that is not on social media. A detail from a recent family gathering, the name of a childhood pet, what you ate together last time. Avoid birthdays and hometowns, which are easy to look up.
- Bring in a second person. Call a spouse, sibling or colleague who can reach the person independently. Many scams collapse the moment one more person is in the loop.
- Delay the money, not the conversation. Even if everything checks out, move money only through your normal process. Urgency is a tactic; real emergencies usually survive a short delay.
- If you already sent money, call the bank immediately. Ask the bank to try to stop or trace the transfer, then report to the police with screenshots, account numbers and call times. Speed matters more than embarrassment.
Write these steps on a card near the phone for elderly parents. In a stressful moment people fall back on habits, not on an article they read last month.
How can you reduce your exposure without disappearing online?
You cannot fully stop someone from making a fake of you, but you can make it harder and less useful. The biggest win is protecting the accounts scammers hijack to start the conversation. Turn on two-factor authentication for Facebook, Zalo, email and banking apps, preferably with an authenticator app or a passkey rather than SMS alone, and review which devices are logged in.
Second, think about how much clear, front-facing video and voice you post publicly. You do not need to delete everything. Setting older albums and long talking videos to friends-only reduces the easy material available to strangers.
Third, tell your contacts what you will never do. A short family message such as “I will never ask for money by video call; if I do, call me back on my number” turns a vague worry into a rule people can follow.
Callback rules for businesses, and the limits of detection tools
For businesses, the same thinking scales up: no payment instruction is approved on the strength of a call alone, however senior the face on the screen. A written callback rule protects junior staff, because they can point to the policy instead of having to doubt the boss.
Detection tools exist and will improve, but they are not yet something an ordinary person can rely on in the middle of a call. For now, the strongest defence is still a slower process and a second channel.
Frequently asked questions
Can a deepfake be done live during a video call?
Yes. Real-time face-swap software can alter a webcam feed during a live call, although quality drops with movement, side angles and good lighting.
Does asking someone to turn sideways still expose a deepfake?
It sometimes helps, because profiles are harder to generate, but newer tools handle it better. Treat it as a hint and still call back on a trusted number.
Is there an app that detects deepfake calls for me?
Some detection tools exist, mostly aimed at businesses and researchers, and their accuracy shifts as new fake methods appear. For everyday calls, a callback routine is more dependable than any detector.
Why do scammers use video if it can be detected?
Because a few seconds of a familiar face removes doubt for most people, especially when they are worried. The call is kept short and blurry on purpose so nobody examines it.
What should I do first if I already transferred money?
Call your bank’s hotline immediately to request a stop or trace, then report to the police with transaction details and screenshots. Do not pay anyone who promises to recover the money for a fee.
Should I stop posting videos of myself?
Not necessarily. Limiting long, clear public videos to friends reduces easy source material, but securing your accounts with two-factor authentication matters more.
