AI Video Camera Movements: Which Ones Work and What They Make People Feel

The camera moves that work best in AI video are the simple ones: a locked-off shot, a slow push-in, a slow pull-out, a slow lateral slide, a rack focus and a gentle handheld. Each one makes the viewer feel something different, so choosing a move is an emotional decision, not a technical one. They are also the moves current models pull off most reliably, as long as you give each shot one move, describe its speed in words, and say where the move ends.

Key takeaways

• Every camera move changes how the viewer feels about what is in the frame. Pick the move for the feeling first.
• Six moves cover almost every ad: locked-off, slow push-in, slow pull-out, slow lateral slide, rack focus and gentle handheld.
• Give each generation one camera move. Prompt guides for Seedance and Kling both warn that stacked moves blur into drift.
• Describe speed with words like slow, steady and gentle, and say where the move ends.
• A push-in is not a zoom. Say which one you want, because they look different.

What is a camera movement in AI video?

A camera movement is how the virtual camera travels, turns or refocuses during a shot. In a real production it is a dolly on tracks, a slider, a crane or an operator's shoulders. In AI video there is no camera at all. The model simulates one, and the words in your prompt tell it what that camera does.

That's why camera language matters so much here. The model can only simulate the move you describe. If you don't describe one, it picks for you, and its pick is usually a slow, aimless drift that means nothing.

Why is every camera move an emotional decision?

Because the camera stands in for the viewer. When the camera moves closer, the viewer leans in. When it pulls away, the viewer is left behind. When it holds perfectly still, the viewer watches like someone who isn't part of the scene.

Yasujiro Ozu is the clearest example of the still camera. Roger Ebert wrote that Ozu "almost never moves his camera" and that in Tokyo Story it moves only once. The camera usually sits about three feet above the floor, roughly the eye level of someone kneeling on a tatami mat. You watch the family from inside the room, calm and patient, and the stillness is the point of view.

Now compare the restaurant scene in The Godfather. Michael sits across from Sollozzo and McCluskey, deciding whether to go through with it, and the camera moves closer and closer to him as he thinks. The screech of a passing elevated train builds over it. Nobody says what he is thinking. The camera tells you something is about to break.

Which camera moves work reliably in AI video?

These are the six moves I use for almost everything. They are the simplest moves to describe, and they are the ones I get back clean most often. For each one: what it makes people feel, where it fits in an ad, and how to prompt it.

1. Locked-off

What it is: the camera doesn't move at all. Only the subject moves.

What it makes people feel: calm, composed, observed. The viewer is a witness who isn't involved. It also puts all the attention on whatever moves inside the frame, which is why it is the strongest opening for an ad. I go into that in How to Hook Viewers in the First Two Seconds of an AI Video Ad.

Best for in ads: product hero shots, premium brands, any shot where the subject's movement is the story.

How to prompt it: "camera locked off on a tripod, zero camera movement." If the model still adds a little drift, the Creative Tim Seedance guide suggests adding "locked tripod, zero camera shake". Saying it twice in different words helps.

2. Slow push-in

What it is: the camera physically travels towards the subject, also called a dolly in.

What it makes people feel: something important is happening inside this person, or about this object. It builds tension and intimacy without anyone saying a word. It's the Godfather move.

Best for in ads: the product reveal, a face at the moment of a decision, the last shot before the logo.

How to prompt it: say the speed, the start and the end. "Very slow, steady push-in from a medium shot to a close-up on her eyes." The seedance.tv prompt guide points out that push-ins often fail because the prompt asks for forward movement but never defines the landing frame. Name it.

3. Slow pull-out

What it is: the camera travels backwards, away from the subject, revealing more of the world around it.

What it makes people feel: context, distance, sometimes loneliness, sometimes scale. The opening of The Godfather is a famous version of this feeling, done with a very slow zoom out rather than a dolly: it starts on Bonasera, the undertaker asking for justice, and widens to reveal Vito Corleone listening in his dark study. The move tells you who has the power before anyone says it.

Best for in ads: endings, showing a product in its world, revealing that the "small" moment was part of something bigger.

How to prompt it: describe what the pull-out reveals, because the model has to invent everything that comes into frame. "Slow, steady pull-out from a close-up of his hands to a wide shot revealing the whole empty beach behind him."

4. Slow lateral slide

What it is: the camera travels sideways, parallel to the subject. Runway calls it a truck, and on set you'd do it on a slider or a dolly track.

What it makes people feel: calm discovery. Objects in the foreground pass faster than the background, and that layered movement makes a flat image feel deep and expensive.

Best for in ads: a row of products, walking past a scene, sliding from behind a foreground object to reveal the subject.

How to prompt it: "slow lateral slide to the right, a wooden post in the foreground passes across the frame." Then check the result for parallax. A Kling camera guide from Atlas Cloud makes a useful test: foreground, subject and background should shift at different rates. If everything moves together, the model gave you a pan, not a slide.

5. Rack focus

What it is: the camera stays still and the focus shifts from one plane to another. Runway's camera terms page describes its purpose as "redirecting attention without moving the camera."

What it makes people feel: a change of attention. The viewer is told: forget that, look at this. It's a reveal without a cut.

Best for in ads: moving from the product to the person using it, or the other way round. It gives you two moments in one generation, which is valuable when most AI clips are a single shot.

How to prompt it: lock the camera, name both planes, and ask for one slow shift. "Camera locked off. Focus starts sharp on the cup in the foreground, then slowly shifts to the man behind it."

6. Gentle handheld

What it is: the slight drift and breathing of an operator holding the camera.

What it makes people feel: presence. Someone is there and it is happening now. Runway describes handheld as "subtle natural shake, slightly imperfect framing" for documentary realism and urgency. It's one of the strongest signals that footage was filmed by a person, which I cover in depth in How to Make AI Video Look Real, Not Like AI.

Best for in ads: lifestyle, UGC style ads, anything that should feel witnessed rather than produced.

How to prompt it: describe the kind of handheld, not just "shaky". "Gentle handheld, subtle sway as if the operator is breathing, small drifts that always recover."

What's the difference between a push-in and a zoom?

This one matters, because models do both and they look different.

A push-in, or dolly in, moves the camera through space. A zoom stays in the same place and changes the lens. Runway puts it simply: "Dolly means the camera physically moves and the background shifts. Zoom means the lens magnifies and it doesn't." Their page lists mixing the two up as a common mistake, because a zoom "flattens perspective when you wanted depth."

In practice: if you want the viewer to feel they are moving into the scene, write "push-in" or "dolly in". If you want the flattening, slightly surveilled feeling of a long lens closing in, write "slow zoom in". Don't leave it to the model to guess.

Which camera moves should you avoid or use with care?

None of these are banned. They just fail more often, or they draw attention to the camera instead of the story.

Fast moves. The Creative Tim Seedance guide calls "fast" the most dangerous word in a Seedance prompt, and says a fast camera with a fast subject in a busy scene almost guarantees jitter. Runway's guide adds that a move with no speed attached tends to come out erratic and too fast by default. Always attach a speed.

Stacked moves. Push-in plus orbit plus tilt plus rack focus in one clip. The seedance.tv guide says the model "often averages those instructions into something soft and indecisive," and the Atlas Cloud Kling guide links stacked orbit, pan, zoom and crane instructions to frame shake. Runway says Gen-4.5 follows prompts well enough that combining moves is encouraged, but still calls one move per clip the reliable fallback. One move per shot.

Orbits. Circling the subject is popular for product shots and the guides list it. The trouble starts when you add other moves on top of it, and big orbits give the model a lot of unseen space to invent. If you use one, keep it small. The Kling guide's own example is only 25 degrees.

Crash zooms. Runway defines it as an extremely fast zoom, for comedy beats and shock reveals. It's a fast move by definition, so it carries every risk above, and in an ad it usually reads as a gimmick.

The dolly zoom. The Vertigo shot, where the subject stays the same size while the background stretches or compresses behind them. It's a dolly and a zoom at once, so test it on its own first.

How do you write a camera move into a prompt?

The same structure works across Seedance, Kling, Runway and Veo:

1. Name one move. Locked-off, push-in, pull-out, lateral slide, rack focus or handheld.
2. Give it a speed. Very slow, slow, steady, gentle. Runway's guide says every movement term works better with a speed or duration attached.
3. Say where it starts and ends. "From a medium shot to a close-up on her eyes." This is the landing frame.
4. Keep the subject's action small. The move is the story. If the subject is also running around, the move gets lost and the model has two hard things to get right.
5. Start from an image when you can. Get the composition right as a still, then animate it. The Atlas Cloud Kling guide recommends image-to-video for exactly this reason, so the layout is fixed before the move begins.

Worked example: the same scene shot five ways

Here's one scene, an ad for a thermos brand, shot with five different moves. Same man, same jetty, same light. Only the camera changes, and with it, what the viewer feels.

Locked-off: calm, observed

Medium wide shot of a man in his fifties in a faded navy wool jumper, sitting at the end of a wooden jetty at dawn, holding a steel thermos cup in both hands. Camera locked off on a tripod, zero camera movement. Steam rises from the cup and drifts left in the breeze. His shoulders rise slowly with a deep breath, then drop. Still grey lake, pale light low on the horizon behind him, fine film grain.

Slow push-in: something is happening inside him

Medium shot of a man in his fifties in a faded navy wool jumper, sitting at the end of a wooden jetty at dawn, holding a steel thermos cup in both hands. Very slow, steady push-in from a medium shot to a close-up on his face, the camera never stops moving. As the camera arrives, his eyes narrow slightly against the light and the corners of his mouth lift. Still grey lake behind him, pale dawn light from the left, fine film grain.

Slow pull-out: he is alone in a big world

Close-up of a man's weathered hands wrapped around a steel thermos cup, steam rising. Slow, steady pull-out to a wide shot revealing him sitting alone at the end of a long wooden jetty, the still grey lake stretching out on both sides, pine forest on the far shore, pale dawn sky. The camera stops on the wide shot and holds. Fine film grain.

Slow lateral slide: calm discovery

Slow lateral slide to the right along the shore at dawn. A weathered wooden post in the foreground passes across the frame, revealing a man in a faded navy wool jumper sitting at the end of a jetty, holding a steel thermos cup. The post moves faster than the man, the far shore moves slowest. The camera stops with the man framed on the right third. Still grey lake, pale light, fine film grain.

Rack focus: from the product to the moment

Close-up of a steel thermos standing on the wooden boards of a jetty in the foreground, sharp, beads of dew on its surface. Behind it, out of focus, a man sits at the end of the jetty looking out over the lake. Camera locked off, no camera movement. Focus slowly shifts from the thermos to the man as he lifts the cup to his mouth and blows across it. The thermos falls soft. Shallow depth of field, pale dawn light.

Notice that every prompt has one move, a speed, and an end point. The man barely does anything, and that's on purpose. The camera carries the feeling.

Camera movement reference table

What mistakes make AI camera work look amateur?

• No move written at all. The model drifts by default, and an aimless drift means nothing.
• Moving the camera because you can. If you can't say what the move makes the viewer feel, use a locked-off shot.
• Stacking moves. One generation, one move.
• No landing frame. The push-in that never arrives anywhere.
• Asking for a zoom when you meant a push-in. The background flattens and the shot loses depth.
• Every shot moving. If every shot pushes in, nothing feels important. Mix moving shots with still ones.

Checklist before you generate

• Can you say in one word what this move should make the viewer feel?
• Is there only one camera move in the prompt?
• Does the move have a speed?
• Does the prompt say where the move ends?
• Is the subject's action small enough that the move stays the story?
• Did you write push-in when you mean push-in and zoom when you mean zoom?
• Are you starting from an image you are already happy with?

FAQ

What is the most reliable camera move in AI video? Locked-off. With no camera movement, the model only has to get the subject's movement right. If drift appears, add "locked tripod, zero camera shake".

How do I stop AI video from making the camera move too fast? Attach a speed to every move, like very slow, slow or steady, and avoid the word "fast" unless you really want it. Keep to one move per shot and keep the scene simple.

Can AI video models do a rack focus? Yes. Rack focus appears in the camera vocabulary of Runway's guide and in Seedance prompt guides. Lock the camera, name the plane that is sharp at the start and the one that is sharp at the end, and ask for one slow shift.

What's the difference between a dolly in and a zoom in AI video? A dolly in moves the camera through space, so foreground and background shift against each other. A zoom only magnifies, so perspective flattens. Write the one you want.

Which camera move is best for a product ad? Start with a locked-off hero shot, use a slow push-in for the reveal, and a rack focus to connect the product with the person. The slow lateral slide is great for showing a range.

Sources

• Runway, "AI camera prompts: how to control camera movement, angles and shots in AI video"
• Creative Tim, Seedance 2.0 prompt cheatsheet
• Seedance.tv, "8 Master Seedance 2.0 Camera Movement Prompts"
• Atlas Cloud, "Kling 4 Controlled Camera Movement: 3 Shots That Stop AI Drift"
• Roger Ebert, "Tokyo Story" (Great Movies)
• No Film School, "4 Filmmaking Lessons from The Godfather's Infamous Restaurant Scene"
• PremiumBeat, "How Invisible Camera Movement Conveys Emotion"
• Wikipedia, "Dolly zoom"

Want to direct AI video like this?

AI Unchained is my community for marketers, brand owners and filmmakers who want to direct AI video, not just prompt it. Every week we break down a film technique and put it into practice, there are three live calls a week, members get our prompt GPTs, and I reply to every piece posted in the Cinema Room the same day.

Join AI Unchained ↗

Running a brand or a content team? I also run in-company AI video workshops, so your team learns to make ads and social content in house.

Book a workshop ↗