Camera Movement Prompts for AI Video: Dolly, Pan, Orbit

Why does your AI video clip sit there like a still photo, or swing around in a way you never asked for? Usually the prompt never told the camera what to do, or it told it three things at once.

Camera movement prompts are the fix. Google's own Veo 3.1 guide is blunt about it: put the cinematography into the prompt, because the model won't guess the move for you. This post covers the four moves you'll use most (dolly, pan, orbit, crane), the wording that tends to work, and the mistakes that waste credits.

The short version

• Name one camera move per shot.
• Say the speed ("slow", "steady").
• Say where the shot starts and what it reveals at the end.
• Put the camera instruction first, in its own sentence.
• Expect to run a few variations. No model gives the same result every time.

The moves, in plain words

Each term below is real film language. Models have seen plenty of it in training, so it's a shorthand they tend to respond to.

Dolly (push in, pull back)

The camera physically travels toward or away from the subject. It's different from a zoom, where the lens changes and the camera stays put. A dolly changes how the background shifts around the subject, which is why it feels more natural.

Kling's camera control guide uses "push in" and "pull back" as its basic terms, and gives patterns like these:

• Camera slowly pushes in on [subject] to emphasize [detail or emotion].
• Camera slowly pulls back from a close-up of a designer ring to reveal a full jewelry display on a clean white table, soft studio light.

The second one is a good template for product videos: start tight, end wide, and say what the wide shot shows.

Pan and tilt

A pan swings the camera left or right from a fixed spot. A tilt does the same up or down. Nothing travels through space, so these are the safest moves when you want the least risk of warping.

Kling's guide suggests: Slow, stable pan left across [scene], no sudden motion. Google's guide lists "slow pan" among its camera movement terms too. The word "slow" matters. Without it, pans can come out rushed.

Orbit (arc shot)

The camera circles the subject. Google's Veo 3.1 guide has a worked example using a 180-degree arc:

The camera performs a smooth 180-degree arc shot, starting with the front-facing view of the singer and circling around her to seamlessly end on the POV shot from behind her on stage.

Notice what makes it work. It states the angle (180 degrees), the start (front-facing) and the end (from behind). "Orbit around the person" gives the model far less to hold on to. Kling's guide lists "orbit slowly" as a move as well.

Orbits are where models struggle most, because the back of the subject has to be invented. Faces, logos and text on products can drift as the camera swings round. Shorter arcs are safer than a full 360.

Crane (rise and descend)

The camera moves vertically on a boom, often while revealing a landscape. Google's guide gives this example:

Crane shot starting low on a lone hiker and ascending high above, revealing they are standing on the edge of a colossal, mist-filled canyon at sunrise, epic fantasy style, awe-inspiring, soft morning light.

Again: a clear start (low on the hiker), a direction (ascending), and a reveal (the canyon). That "reveal" structure shows up in nearly every good camera prompt.

A prompt structure you can reuse

Google recommends this order for Veo 3.1: cinematography, then subject, action, context, and style and ambiance. You can borrow that for any model. For example:

Slow dolly in. A barista pours latte art at a wooden counter, steam rising, small cafe in early morning, warm window light, shallow depth of field.

Compare that with a barista makes coffee, cinematic camera movement. The second one gives the model nothing to act on. Kling's guide says the same thing: use clear direction words rather than asking for something "cinematic".

Google also notes you can pair the move with a lens or focus cue, such as shallow depth of field or a wide-angle lens, for more control. And keep clip length in mind. The guide says Veo 3.1 clips run 4, 6 or 8 seconds, so a move that needs a long runway (a full orbit plus a crane) won't fit comfortably.

Mistakes that burn credits

1. Stacking moves. "Dolly in while panning left and orbiting" is asking for trouble. Kling's guide recommends one main camera move per shot. Need more? Make two clips and cut them together.
2. Contradicting yourself. Kling's guide specifically warns against things like "static camera with fast push-in". Pick one.
3. Changing the subject at the same time. The same guide advises against combining a push-in with several subject changes at once. If the person is running, turning and the camera is moving, something will give.
4. Leaving speed out. Add "slow", "steady" or "smooth".
5. Skipping the end state. Say what the camera should be looking at when it stops.
6. Expecting exact repeats. Kling's guide calls results "probabilistic" and says testing variations is necessary. That's honest advice for every video model.

Model differences to expect

Not every model is documented equally. Google and Kling both publish camera guidance. For Wan, I checked the official Wan2.2 repository: its README describes training on aesthetic data labelled for lighting, composition, contrast and color tone, but it doesn't list a camera control feature. So if you see Wan camera tips elsewhere, treat them as community experience rather than a guarantee. Test them yourself.

The practical takeaway: write the same camera prompt for two or three models and compare. Some will follow a crane move well and ignore an orbit. Others do the opposite.

Testing camera moves in Mediafy Studio

In Mediafy Studio, choose Video, pick a model (models such as Kling v3, Veo 3.1 and MiniMax H3 Max are offered, and the lineup changes over time), and paste your prompt. The credit cost shows on the Generate button before you run it, and failed generations are refunded automatically. Costs differ by model, so check before you run a batch.

A cheap workflow:

1. Type a short prompt like slow dolly in on a ceramic mug.
2. Press Enhance to get a fuller version, then check that your camera instruction is still first and still a single move. Edit it back if not.
3. Run the same prompt on two models and compare.
4. Save the winner in My Prompts so you can reuse the wording.

If you're stuck on phrasing, Mediafy Chat can help brainstorm prompts. And if you're new to the Studio, the first image tutorial walks through the basics. If you want a more structured path through AI creation, there's also the AI Creator Lab course.

What camera prompts can't do

They can't guarantee a precise path. You can't ask for "exactly 40 centimetres of travel", and you can't count on the same move twice. Characters can change face between frames during a big orbit. Text on packaging can wobble. Treat camera words as strong hints, then pick the best take.

When you need the same look across clips, start from a strong still and animate it with a simple move. Simple moves, such as a slow push-in or a gentle pan, are the most dependable ones to start with.

Ready to try a few? Sign in at /signin and run your first dolly.

Sources

• Google Cloud, "The ultimate prompting guide for Veo 3.1": https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-veo-3-1
• Kling AI, camera control video guide (Kling blog): https://kling.ai/ar/blog/kling-ai-camera-control-video-guide
• Wan-Video, Wan2.2 repository README (GitHub): https://github.com/Wan-Video/Wan2.2