Camera Movement Prompts for AI Video: The Complete Visual Guide

Why Camera Language Is the Highest-Leverage Prompt Ingredient

Most people writing AI video prompts spend all their effort on the subject and the scene — what the person looks like, what they're wearing, where they're standing — and leave the camera to chance. That's backwards. The model already has decent instincts for subjects and environments. What it doesn't have is a director standing over its shoulder saying "no, pan left here, not a static shot."

Naming an actual camera movement — not "make it cinematic," but "camera dollies in while racking focus from background to subject" — is the difference between a video that looks like a screen recording and one that looks shot. It costs you one sentence and changes the entire feel of the output.

We built out a camera movements library with 42 named moves, each with a short reference video and a ready-to-paste prompt line, because this is the part of prompting most people skip. Below is a tour of the seven families, how to combine them, and where you can already see them baked into a Starrd template.

The 7 Families

Pan & Tilt — the camera rotates in place

The camera doesn't travel anywhere here; it just turns. Pan sweeps the frame horizontally at a fixed position, left or right — good for revealing what's beside your subject or following lateral motion without a cut. Tilt does the vertical version, and tilting up is the move that sells scale: start on someone's feet, tilt up, end on a building towering over them.

If you want urgency instead of smoothness, whip pan snaps between two subjects almost instantly — use it as an accent, not a foundation. It only works if there's genuinely a "before" and "after" subject to connect.

Zoom & Lens — the lens does the moving, not the camera

This family covers focal-length tricks rather than physical movement. Zoom in and crash zoom both push toward the subject, but crash zoom does it fast and jarring — a comedic or shock beat, not a dramatic one. Rack focus keeps the camera and framing completely still and just shifts what's sharp, which is the cheapest way to redirect attention mid-shot without any motion at all.

The showcase move here is the dolly zoom (vertigo) — Hitchcock's trick where the camera physically moves one direction while the lens zooms the other, so the subject stays the same size but the background stretches or compresses around them. It reads as disorientation or dread almost automatically, which is why it's overused in horror — use it once, not per-scene.

Dolly & Track — the whole camera travels

This is the workhorse family. Dolly in is the single most reliable move for building tension toward a reveal or an emotional beat. Tracking shot follows alongside a moving subject and is your default for anything that involves someone walking or running — pair a tracking camera with a moving subject, never an orbiting one, or the model gets confused about which motion to prioritize.

Truck moves the whole camera sideways rather than rotating it — useful when you want the background to shift in parallax rather than just sweep past, which reads as more "real camera on a rig" than a pan does.

Physical Moves — arcs, orbits, and full-body camera motion

Orbit circles the subject, left or right, and is the go-to for hero moments — someone standing still while the world rotates around them. The three-sixty orbit is the full commitment version of the same idea, usually reserved for a single climactic beat rather than something you'd use twice in one video.

Low-angle rise starts near the ground looking up and climbs — power-pose energy, the shot that makes a subject look larger than life. Arc shot curves partway around the subject instead of committing to a full orbit, which reads as more natural; use it when you want a partial reveal rather than a full circle.

Drone & Crane — elevated, sweeping aerial moves

Crane moves vertically on a boom rather than tilting in place, which gives you actual parallax as the camera rises or descends — grander than a tilt, because the camera position itself is changing, not just its angle. Drone flyover is the establishing-shot move: wide aerial coverage of a whole scene before you cut down to ground level.

Aerial orbit combines the elevation of a drone with the circling motion of a physical orbit — reserve it for scenes that are genuinely worth a full aerial reveal, like a stadium or a skyline, since it reads as expensive and can feel out of place on a small, intimate scene.

Human Camera — handheld and body-worn energy

Sometimes the point is imperfection. Handheld follow adds natural shake while tracking a subject, which reads as documentary or found-footage rather than staged — great for chase scenes or anything that should feel visceral instead of polished. POV walk puts the viewer directly in the subject's shoes, first-person, no visible camera operator at all.

Snorricam is the specialty version — a rig mounted to the subject's body so they stay locked in frame while the world spins and shakes around them. It's a strong, recognizable look (Requiem for a Dream, Fight Club) and it only works if the disorientation is the point of the scene.

Specials — signature cinematic effects

Bullet time freezes the action while the camera continues moving around the subject — the Matrix move, and it's exactly as expensive-feeling as you'd expect, so save it for a genuine climax. Speed ramp shifts from real-time into slow motion (or back) mid-shot, which is the single best way to punch up an impact moment without needing a cut. Hyperlapse compresses a long journey or the passage of time into a few seconds of blurred, accelerated motion — the "time is passing" shorthand.

How to Combine Moves

A single named move will usually beat a paragraph of vague camera language, but stacking two complementary moves in sequence — not simultaneously — is how you get results that feel directed rather than just shot. The camera movements library includes a combos category for exactly this: pre-built pairings that are proven to read well together, like a crane-down-push-in (descend from an aerial establishing view, then push straight into a close-up) or an orbit-slow-zoom (circle the subject while the lens tightens at the same time, compounding the drama of both moves).

The rule that keeps combos from turning into chaos: pick one move that establishes the frame and a second move that resolves it, and describe them as sequential beats — "camera cranes down from above the arena, then pushes in tight on her face as she looks up" — rather than asking for everything at once. Two clean, sequenced moves will consistently outperform three or four stacked simultaneously.

How Starrd Templates Already Encode These Moves

You don't have to write any of this from scratch to see it working. A handful of Starrd's published templates are built around a specific camera move as their core beat:

• Skyfall leans entirely on an extreme slow-motion falling shot — the "10% speed" ramp is the single line that makes the whole template work, a direct descendant of the speed ramp family.
• Red Carpet uses paparazzi-style handheld energy for the arrival beat, then settles into steadier coverage once the subject is posing — the same handheld-to-locked contrast that handheld follow describes.
• Pitch Invader (World Cup) uses broadcast-distance tracking and crowd-level camera work to sell the "caught on the pitch" moment, borrowing from the same tracking shot logic covered above.

If you're writing your own prompts, the fastest way to get a feel for any of these moves is to watch the reference clip on its camera movements page, then copy the prompt line and swap in your own subject and scene. The camera language is reusable across almost any first frame — that's the whole point of keeping it separate from the subject description.