Why Your AI Video Looks Like AI Video

Most people can't explain why a video feels generated. They just know it does, and usually within the first couple of seconds. They don't stop to analyse it. Something in the frame breaks a rule their eyes have followed their whole lives, and trust goes before they can name the reason.

For a brand, that moment costs more than one view. A viewer who clocks an ad as AI starts asking what else the brand cut corners on. So it's worth knowing exactly what they're reacting to. Below are the six tells we see most often in AI video reels, with the reason each one happens. The fixes are a separate conversation. This piece is about learning to see the problem.

1. The face that drifts between shots

Shot one: a woman in her thirties, strong jaw, a small scar through one eyebrow. Shot two, same scene: the jaw is softer, the eyes a little further apart, and the scar is gone. Each frame looks fine. The cut is where she stops being the same person.

Why it happens. Every shot is a separate generation. The model has no memory of the person it drew thirty seconds ago. It rebuilds a plausible face from whatever it was given each time, and "plausible" leaves room for a lot of small variations. Humans are extremely good at recognising faces, so we notice those variations long before we could point to any one of them.

2. Hands and objects that ignore physics

A hand closes around a coffee cup and two fingers pass through the handle. A phone changes shape as it's picked up. Someone pours a drink and the liquid never reaches the glass.

Why it happens. Video models learn how things look from frame to frame. They have no notion of how an object is built or what it weighs. Contact is the hardest case: two objects touching, one blocking the other, fingers bending around a curve. Hands make it worse because they're small, have a lot of joints, and are half hidden in much of the footage the models learn from.

3. The room that changes between cuts

The wide shot has a window on the left and a plant by the door. In the close-up, the window is behind the actor and the plant has turned into a lamp. The kitchen counter goes from marble to wood between two lines of dialogue.

Why it happens. There is no set. For every shot, the model paints a new image of a room that fits the description, with no fixed room behind it. Background detail is also where the model spends the least effort, because the subject gets priority. Viewers rarely notice the lamp consciously, but they register that the space stopped making sense.

4. Dead eyes on the emotional peak

This is the line the whole ad was built for. The mouth moves, the brows lift, the face does everything a surprised face should do, and the eyes stay flat. It looks like an actor reading the line without feeling it.

Why it happens. Models reproduce the average of what they've seen, and extreme emotion is rare compared with neutral faces. The life in a real person's eyes comes from tiny, fast signals: small shifts of gaze, the tension around the lids, the timing of a blink. Those are the first things averaging smooths away. It hits hardest at the emotional peak because that's exactly where the viewer looks straight at the eyes.

5. Light that jumps between cuts

In the wide shot, the sun comes through a window on the left. In the reverse angle, the key light is on the right, the skin has gone from warm to cool, and the shadows point away from a window that should be lighting them.

Why it happens. A new shot doesn't inherit the lighting setup from the last one. Each generation works out its own light from the prompt and from its idea of a good-looking image. Every shot can be beautiful and the sequence still falls apart, because real light obeys one source across a whole scene. We cover the mechanics in Lighting a Scene Shot to Shot.

6. Motion without weight

People glide more than they walk. A car takes a corner with no body roll. Hair and fabric drift as if the scene were underwater. The camera moves with a smoothness no operator, dolly or drone would produce.

Why it happens. A video model predicts the next plausible frame. It doesn't simulate mass, force or friction. The result is movement that is smooth but lacks the cues that tell us something is heavy: the push before a step, the lag in a turn, the settle after a stop. We feel that absence even when we can't describe it. Many models also lean toward slow, drifting motion because it's easier to keep coherent than fast motion with hard stops.

How to use this when you judge a reel

You don't need a technical background to evaluate an AI video vendor. You need to watch in the right places.

Where the six tells show up

Three of the six only show up between shots

Where the six tells show up

Tell	Where it shows up
1. The face that drifts between shots	At the cut
2. Hands and objects that ignore physics	Inside a shot
3. The room that changes between cuts	At the cut
4. Dead eyes on the emotional peak	Inside a shot
5. Light that jumps between cuts	At the cut
6. Motion without weight	Inside a shot

• Watch the cuts. Three of these six tells only show up between shots. A reel built from single hero shots hides them completely. Ask to see one continuous sequence of three or more connected shots.
• Pause on the hands. Find a moment where a character touches or holds something. If the reel avoids hands entirely, that tells you something too.
• Look at the eyes at the peak. Scrub to the most emotional moment and watch the eyes only.
• Check the room. Pick one object in the wide shot and look for it in the next angle.
• Feel the weight. Watch a body or a vehicle stop. Real things settle.

No production is perfect on all six every time. Still, each of these tells is predictable, which means a team either manages it on purpose or leaves it to chance. A good reel shows it was managed.