AI video camera movements: the shot language models understand
Shot sizes, angles, lenses, moves and light: the camera terms to write into video prompts, with two reference tables and three ready-to-use examples.
AI video camera movements come from writing the language of the set: push-in, pan, tracking shot, crane, orbit, handheld. Models also understand shot sizes, angles, lenses and lighting, but they respond best to precise instructions: camera before subject, one move per shot, a stated speed, and what should stay still.
Below are the reference tables we use in the studio, with the terms to write into a prompt, what each choice says, and what the tools offer as of November 2026 when text alone isn’t enough.
Why the camera has to be written, not left to the model
A video model fills every gap with the most probable choice. If you don’t specify framing and movement, it decides for you, and the result tends to look flat or generic. That’s the observation in an each::labs guide to Veo 3 (February 2026), which also recommends leading with the camera, because the model gives more weight to what comes early in the prompt.
The model makers say much the same. In the Veo tips reported by 9to5Google in 2025, Google lists camera angle and movement among the key elements of a prompt, alongside subject, action, setting, lighting and style. Runway’s Gen-4 prompting guide asks for concrete physical actions rather than abstract ideas, and one scene per generation; for Gen-4.5 it has published a dedicated guide to camera terms.
The rule is the same as on set: every shot has a size, a height, a lens and a movement, and each of them tells part of the story. Film a scene from below and the character looms; film it from above and they shrink.
Shot sizes, angles and lenses: the reference table
The basic sizes, with the term to write and the job each one does.
| Shot | In the prompt | What it says |
|---|---|---|
| Extreme wide | extreme wide shot | The place dominates, the person is a dot |
| Wide | wide shot, establishing shot | Where we are |
| Full | full shot | Body, posture, wardrobe |
| Medium wide | medium wide shot, cowboy shot | Action and gesture |
| Medium | medium shot | Dialogue, presence |
| Close-up | close-up | Emotion |
| Extreme close-up | extreme close-up | One detail of the face |
| Insert | insert, macro shot | Object, product, texture |
| Over-the-shoulder | over-the-shoulder shot | The relationship between two people |
| Point of view | POV shot | What the character sees |
Angle and height
Camera height shifts the balance of power. Eye level is neutral; a low angle makes the subject bigger; a high angle makes them smaller; a top-down or bird’s-eye view turns the scene into a map; a Dutch angle, tilted off the horizontal, signals unease. Pick one per shot and keep it consistent within a scene: each::labs also advises holding camera height and lighting steady across a sequence.
Lenses and focus
Focal length works as a style instruction. A wide-angle lens (24mm or below) opens up space and distorts the edges; a 35mm gives a natural view; an 85mm compresses the background and brings the face closer; a telephoto flattens the planes; a macro gets inside the material. For depth of field: shallow focus to separate the subject, deep focus to keep everything sharp, rack focus to shift attention from one plane to another.
Some platforms turn these choices into parameters. Higgsfield and Adobe are Kult Studios partners. According to a Higgsfield page updated in November 2026, Cinema Studio offers camera bodies such as 35mm Film, 8mm Film and DV Camcorder, lenses such as Clean Sharp, Anamorphic and Vintage Anamorphic, and apertures of f/1.4, f/4 and f/11; in version 4.0 the lens is chosen by character and aperture rather than in millimetres.
AI video camera movements: what to ask for and how
Movement demands more precision than anything else in the prompt.
| Move | In the prompt | When to use it |
|---|---|---|
| Static | static shot, locked-off camera | Let the action speak |
| Pan | pan left, pan right | Reveal the space |
| Tilt | tilt up, tilt down | Reveal height or a detail |
| Dolly in or out | push-in, pull-back | Move closer to a thought, or isolate |
| Lateral move | tracking shot, truck left | Travel with someone moving |
| Crane | crane up, crane down | Open or close a scene |
| Orbit | orbit, arc shot | Circle a subject or product |
| Handheld | handheld camera | Urgency, documentary feel |
| Aerial | aerial shot, drone shot | Scale and geography |
| Vertigo effect | dolly zoom | Disorientation, a reveal |
Speed, and what stays still
Naming the move isn’t enough. Runware’s guide to Runway Gen-4.5 treats pacing as a second dial: slow, gradual or steady for a measured move, subtle or gentle for an almost imperceptible one, rapid or fast for urgency, sudden for a jolt. Without a speed word, the move tends to come out average.
The same guide treats camera and subject as two independent channels and recommends saying what holds still, such as “the camera holds completely still”, because otherwise the model tends to animate both. And always phrase it positively: Runway warns that “Negative phrasing is not supported” and can produce the opposite result, so write “static camera” rather than “no camera movement”.
Common mistakes
- Zoom instead of dolly. A zoom magnifies without changing perspective; a push-in moves the camera and changes the relationship between subject and background. Write the one you actually want.
- Three moves in five seconds. Pan, then crane, then orbit usually ends up muddled. One move per shot; the edit does the rest.
- Adjectives instead of terms. “Dynamic” or “epic” tells the camera nothing; “low-angle tracking shot, fast” does.
- Re-describing the image. In image-to-video, Runway advises describing the motion, not what’s already in the start frame.
The vocabulary of light
Light is half the shot. Describe it the way a cinematographer would: when, what quality, from where, how much contrast, what colour temperature.
- Time: golden hour, blue hour, overcast daylight, night with practical lights.
- Quality: hard light for crisp shadows, soft light for a gentle face, volumetric light with haze.
- Direction: backlight, rim light on the outline, side light, top light.
- Contrast: high-key for bright, clean scenes, low-key and chiaroscuro for dramatic ones.
- Temperature: warm tungsten, cool daylight, neon, mixed lighting.
The examples PiAPI draws from Google’s Veo best practices point the same way: warm golden-hour light, cool blue tones, soft morning fog. Choose one lighting set-up per scene and repeat it in the same words in every prompt: across a run of clips, continuity often breaks on the light before it breaks on the face.
Beyond the prompt: storyboards, keyframes and presets
When text isn’t enough, the tools offer explicit controls. Kling Video 3.0 Omni, launched in February 2026, has a storyboard where you set duration, framing, perspective, content and camera movement for each shot (Bernama). Introducing version 4.0, Kling describes up to ten keyframes to fix the state of the image at different points on the timeline.
Higgsfield’s Cinema Studio 4.0 offers more than 30 preset camera moves and over 50 colour palettes. In Firefly Boards, a July 2026 Adobe tutorial animates a storyboard panel with Kling 2.5 Turbo by adding a single camera instruction: a slow zoom towards the product.
The most reliable control is still the image. An approved start frame locks size, height and lens, leaving the prompt to handle only the movement. That’s why camera moves are decided on the storyboard, not at generation time.
Three example prompts
The structure never changes: camera, subject and action, light, what stays still.
Static macro shot at eye level, shallow focus. A drop of condensation slides down a cold glass bottle on a marble counter. Soft backlight, cool morning tones. The camera holds completely still.
Medium close-up on an 85mm lens, slow push-in. A woman in a red coat stops at a pedestrian crossing and looks up. Golden hour, warm rim light on her hair, the street softly out of focus behind her.
Low-angle tracking shot, wide-angle lens, fast handheld movement. A runner sprints along a wet track at night, stadium lights flaring behind her.
The camera is a storytelling choice
Knowing the terms doesn’t make you a director. You need to know why this shot is a close-up rather than a medium, and why the camera pushes in instead of holding. Models keep getting better at executing; deciding is still human work.
That’s the point of “Powered by AI. Directed by humans.”: at Kult Studios, sizes and moves are decided in pre-production, and in production every shot goes to the engine best suited to it. To see why a prompt alone won’t do it, read our piece on direction; for a project, get in touch.
Frequently asked questions
Should camera directions in a video prompt be technical or descriptive?
Technical, wherever possible. A named move such as push-in or crane up, plus a speed word, is far more reliable than a description like a dramatic sweeping shot.
Why does the model ignore the camera move I asked for?
Usually the prompt asks for several moves at once, uses a negative, or contradicts the start frame. Put the camera first, ask for a single move with a speed, and try again.
Can you choose a lens in millimetres for AI video?
As a style instruction, yes: 35mm or 85mm changes how space and background render. It isn’t real optics, though, and some platforms use lens and aperture presets instead of millimetres.
Sources
- Gen-4 Video Prompting Guide · Runway Help Center, 2026-11
- Gen-4.5: prompting guides and camera terms · Runway Help Center, 2026-11
- Runway Gen-4.5: directing motion · Runware Docs, 2026-11
- Google shares tips for writing better Veo video prompts in Gemini · 9to5Google, 2025-05-03
- Veo 3 API: A Step-by-Step Guide to Writing Good Prompts Using Google API Best Practices · PiAPI, 2025-12-04
- Structuring Veo 3 prompts for better motion control · each::labs, 2026-02-05
- Best AI video tools with real camera and lens controls for directors · Higgsfield, 2026-11-05
- Cinema Studio 4.0 · Higgsfield, 2026-11
- Kling AI launches the Kling 3.0 series (press release) · Bernama MREM, 2026-02-06
- Kling 4.0 AI video model · Kling AI, 2026-09
- Create a commercial storyboard with Firefly Boards (tutorial) · Adobe, 2026-07-07
Keep reading.
AI storyboards and previz: from script to animatic, before the video
From script to boards to animatic: how an AI storyboard is made, which tools to use as of November 2026 and what to get approved before generating the video.
AI music video: how to make one, from the track to the final cut
From the first listen to the cut on the beat: how an AI music video is built in 2026, and what to agree with the artist and label first.
AI Christmas ads in 2026: who uses AI, who refuses it and why
Prada draws by hand, Levi’s goes back to claymation, Dior faces AI accusations: Christmas 2026 shows what audiences punish and accept from AI.