You write a detailed prompt about a lighthouse on a cliff at dusk. The scene comes out right, but the camera drifts left, zooms a little, then shakes for no reason. The picture is fine. The clip still feels amateur.
The cause is usually simple. The prompt described what to show and said nothing about how to film it, so the model picked a camera move for you.
This guide gives you the camera vocabulary that AI video models respond to: shot sizes, angles, and movements, each with a prompt you can copy.
Why Camera Language Matters in a Prompt
The same subject reads very differently depending on the camera:
- A slow push-in on a face says "pay attention, something is changing inside this person."
- A pull-out from the same face says "look how small and alone they are."
- A handheld shot of that face says "this is happening right now, and it is not calm."
Clips are also short. On veo4.dev, Veo 4 makes clips of 4, 6, or 8 seconds, Kling and Seedance 2.0 offer 5 or 10 seconds, and Hailuo 2 makes 6-second clips at 1080p or 6 or 10 seconds at 512p and 768p. That leaves room for one clear camera idea, so the camera line deserves as much care as the subject line.
Part 1: Shot Sizes
Shot size is how much of the subject fits in the frame. State it at the start of the prompt.
| Shot size | What the frame shows | Best for |
|---|---|---|
| Extreme wide shot | A vast place, the subject is tiny | Scale, isolation, opening shots |
| Wide shot | The full subject and the surroundings | Action, location, context |
| Medium shot | The subject from the waist up | Dialogue, product demos, presenters |
| Close-up | A face or a single object | Emotion, detail, texture |
| Extreme close-up | One small detail, such as an eye or a watch dial | Drama, reveals, product macro shots |
Extreme wide shot of a lone hiker crossing a white salt flat at sunrise, the horizon perfectly flat, long soft shadows, static camera.
Close-up of an old fisherman's weathered hands tying a knot in wet rope, shallow depth of field, warm morning light.
Part 2: Camera Angles
Angle is where the camera sits relative to the subject, and it changes how the viewer feels about them.
Eye level
Neutral and honest. Use it for interviews, presenters, and everyday scenes.
Low angle
The camera looks up. The subject seems strong, tall, or threatening.
Low-angle shot of a basketball player standing under the hoop, looking down at the camera, stadium lights flaring behind him, slow push-in.
High angle
The camera looks down. The subject seems small, tired, or vulnerable.
Overhead (top-down)
The camera looks straight down. It works well for food, desks, board games, and flat lays.
Top-down overhead shot of hands arranging fresh sushi on a dark slate board, clean studio lighting, static camera.
Part 3: Camera Movements
Each prompt below names the move, the direction, and the speed. Leave out the speed and the model picks one for you.
Static shot
The camera does not move; only the subject does. It looks deliberate and calm, and it is the safest choice when the action is complex.
Static shot, eye level. A barista pours steamed milk into a cup, drawing a leaf pattern in the foam. Soft window light, shallow depth of field.
On veo4.dev, the Seedance 2.0 settings also include a Fixed camera option. Turn it on when you want the frame locked, and keep "static shot" in the prompt so both signals agree.
Push-in (dolly in)
The camera moves physically toward the subject. It builds focus and tension.
Slow dolly push-in toward a young woman reading a letter at a kitchen table, her expression changing from calm to shock, warm evening light.
Pull-out (dolly out)
The camera moves away from the subject and reveals the surroundings. It is good for endings and reveals.
Slow pull-out from a close-up of a single red tent to a wide shot of an enormous snowy mountain valley, the tent becoming a tiny dot.
Pan
The camera stays in place and turns left or right, to follow movement or reveal a scene sideways.
Slow pan from left to right across a crowded night market, steam rising from food stalls, paper lanterns glowing, people walking past.
Tilt
The camera stays in one place and turns up or down. A tilt up is a classic way to show height.
Slow tilt up from the base of a giant redwood tree to its canopy, sunlight breaking through the high branches, morning mist.
Tracking shot
The camera moves alongside the subject at the same speed, giving energy without chaos.
Tracking shot moving alongside a cyclist riding along a coastal road, the camera at the same speed, ocean on the right, golden hour light.
Orbit (arc shot)
The camera circles the subject. It feels heroic or premium, which makes it a favorite for product shots.
Slow 180-degree orbit around a glass perfume bottle on a black reflective surface, soft rim light, tiny particles floating in the air.
Crane shot
The camera rises or falls vertically, often while it looks at the subject. A crane up at the end of a scene feels like a closing statement.
Crane shot rising from a couple dancing in an empty town square to a high view of the whole square at night, string lights overhead.
Handheld
A slight, natural shake, as if a person holds the camera. It feels raw, documentary, and urgent.
Handheld camera, documentary style. A street musician plays saxophone on a subway platform as a train arrives, people rushing past, natural light.
Drone shot (aerial)
A high, smooth flying view. It is ideal for landscapes, cities, roads, and coastlines.
Aerial drone shot flying forward over a winding mountain road through autumn forest, a single red car driving through the curves, soft morning fog.
Zoom
A zoom changes the lens, not the camera position, so the background does not shift as in a dolly. A fast "crash zoom" feels comic or dramatic.
Fast crash zoom into the face of a surprised cat as a cucumber appears next to it, comedic timing, bright kitchen.
Rack focus
The camera does not move, but the focus shifts from one subject to another. It guides the viewer's eye inside a static frame.
Static close-up, rack focus from raindrops on a window in the foreground to a woman standing outside under a streetlight in the background.
A Simple Template for the Camera Line
Use this order and you will rarely get a random camera:
[Shot size], [angle]. [Camera move] [direction] [speed]. [Subject and action]. [Light and mood].
Here is the template filled in:
Medium shot, low angle. Slow push-in. A chef flips vegetables in a flaming wok, sparks rising. Dark kitchen, warm orange firelight.
You do not need every slot. If the camera matters, put it in the first sentence so it is not buried under scene detail.
Common Mistakes and How to Fix Them
Stacking several moves in one short clip
"Pan left, then tilt up, then orbit, then crane down" does not fit in 5 or 8 seconds. The model blends it all into a wobbly path. Use one move per clip. For two moves, generate two clips and cut them together.
One combination often works: a primary move plus a much smaller secondary one, such as "slow orbit with a slight rise".
Forgetting the speed
"Push-in" alone can come out fast and jumpy. Add "slow", "gentle", or "smooth" unless you want energy.
Fighting the action
If a dancer spins fast while the camera orbits fast, the motion becomes noise. When the subject moves a lot, keep the camera calm.
Contradicting yourself
"Static shot, handheld" gives the model two opposite instructions. Reread the camera line and remove any word that disagrees with the rest.
Describing the camera, but not the subject's motion
A camera move over a frozen scene looks like a slideshow. Give the subject something to do: hair in the wind, steam rising.
Ignoring the frame shape
Wide moves like pans and drone shots read best in 16:9. In a vertical 9:16 clip, a tilt or a crane uses the tall frame better than a pan does. Veo 4, Kling, and Seedance 2.0 all offer both 16:9 and 9:16 on veo4.dev, so choose the ratio before you write the move. For more on vertical framing, read Making vertical AI videos for TikTok, Reels, and YouTube Shorts.
If the camera still misbehaves after these fixes, Why your AI video looks wrong covers more problems and their prompt fixes. When you film the same person across several shots, also read How to keep the same character consistent across many AI video clips.
FAQ
Which camera move is the most reliable for AI video?
A static shot and a slow push-in are the most predictable. Start with them, then try orbits, cranes, and drone shots.
What is the difference between a dolly and a zoom?
A dolly moves the camera through space, so the background shifts relative to the subject. A zoom only changes the lens. Write "dolly push-in" for depth and "zoom in" for a flatter effect.
Can I use a negative prompt to stop unwanted camera shake?
On veo4.dev, the Kling model passes a negative prompt to the model, so "camera shake, jitter, sudden zoom" there can help. For every model, a clear positive instruction such as "static shot, smooth and stable" is the more dependable fix.
How do I get camera moves in image-to-video?
Add your image as the first frame, then describe only the motion: the camera move, its speed, and what the subject does. The image already shows the scene, so repeating it wastes the prompt.
How many camera moves should one clip have?
One. With clips of 4 to 10 seconds, one clear move gives the cleanest result.
Try It Now
Pick one shot from this guide, copy the prompt, and change only the subject. Then keep the subject, change only the camera move, and compare the two clips side by side.
Open the AI video generator to start, or try camera moves on the Veo 4 page. For product orbits and push-ins, the workflow in A practical workflow for AI product video ads puts these moves to work. Plans and credit options are on the pricing page.