A real face is never still for long. It twitches, resets, contradicts itself. When I watch a suspected deepfake, the first thing that gives it away is the opposite — a face that holds one expression as if it has been pasted on. That is exactly what happened in the manipulated clip of Rashmika Mandanna that did the rounds. Watch the moment she enters the frame: her facial expression is fixed, and it stays fixed. There is no change in the structure, no shift in the muscles, no movement that answers what is happening around her. That sameness is the tell.

When I'm checking whether a face in a video is real, I divide it into three zones and study each one in turn.

The three zones I watch

  • The eyes and eyebrows. This is where liveliness lives. A genuine face shows micro-shifts here constantly — a brow that lifts a fraction, eyes that narrow or widen with the moment. If this region stays flat while everything else changes, be suspicious.
  • The cheeks and the nasolabial area. The fold that runs from the nose down past the mouth deepens and softens with real emotion. Deepfakes often render this zone too smoothly, or fail to move it in time with a smile or a frown.
  • The mouth and lips. The most obvious zone, and the one synthetic models still struggle with. Lip movement that doesn't quite track the words, or a mouth that holds a shape while the rest of the face changes, points to manipulation.

The real test: are the expressions in sync?

A single frozen zone might be a still moment. The stronger signal is when the three zones stop working together. In a genuine human face, emotion is whole — the eyes, cheeks and mouth move as one system, in the same instant, with the same intensity. In a deepfake, you often see them drift apart.

I noticed this clearly in a clip where Deepika Padukone's face had been mapped onto a fast action scene. The scene called for aggression — she is in a launch position, moving quickly, pushing someone aside, dropping her hand and eyes. The body is doing all of that. The face is not. There is no flash of effort, no surge of intensity to match the movement. Her expression stays the same through a sequence that should have torn across her features. The eyes, the cheeks, the mouth — none of them were really in sync with each other or with the moment.

Why this works

Real emotion fires across the whole face almost simultaneously. We can't easily fake that synchrony, and neither, for now, can most generative models. They are good at the geometry of a face and weaker at the choreography of feeling. So I don't look for one dramatic glitch. I look for coherence — whether the brow, the cheeks and the mouth are telling the same story at the same time.

None of this is a guarantee. The technology is improving fast, and a clean clip can still be fake while a poorly lit real video can look uncanny. Treat these cues as a pattern to weigh, not a verdict. But the next time a forwarded video makes you uneasy, slow it down. Watch the eyes, the cheeks, the mouth. Ask whether they are moving as one face would. More often than not, the answer is sitting right there on the screen, refusing to change.