AI video camera movement prompts: choose the move before the words.
Start with one camera move, a direction, a pace, and a visible ending. Describe the subject's action separately. A pan, a push-in, and a zoom can all bring attention to something, but they ask for different changes to the view.
1. Choose what should change in the view
Use camera terms to explain a visual intention. In generated footage, these words are requests; they are not exact physical controls or a guarantee that the model will reproduce real camera geometry. Runway's current camera terminology guide distinguishes the movements below. Its examples are for Gen-4.5; check your own provider's guidance before transferring a prompt.
| Move | Physical meaning | Use it to request |
|---|---|---|
| Pan | Rotate left or right from a fixed position. | A horizontal reveal from one part of a scene to another. |
| Tilt | Rotate up or down from a fixed position. | A vertical reveal, such as moving attention from a doorway to its sign. |
| Dolly / push-in / pull-back | Move the camera toward or away from the subject. | An approach to a detail, or a retreat that reveals its surroundings. |
| Zoom | Change lens focal length without moving the camera. | Tighter or wider framing from the same viewpoint. |
| Tracking | Move with a moving subject. | Follow a walker or cyclist while keeping them in view. |
| Orbit / arc | Travel around a subject; an arc covers part of that path. | A changing view of the subject's sides and background. |
If the subject must remain easy to inspect, start by asking for a small move. A full orbit from one reference photo asks the generator to invent surfaces and surroundings that the image never showed. Decide whether that invention is acceptable before using the result.
For a reveal, name what should be visible at the end. “Pan right” gives a direction, but “pan right to finish on the doorway” gives you an ending to check. For tracking, specify the subject's direction as well as how the camera follows it.
2. Separate the camera, subject, and ending
For image-to-video, let the reference establish the opening scene. Runway's current image-to-video prompting guide recommends concentrating on motion and starting simply. You can identify an object briefly without rewriting every color and texture already visible in the photo.
- Camera: pick one movement and direction.
- Pace and extent: describe a slow, small approach or a short sideways move.
- Subject: say whether it stays in place or performs one action.
- Ending: name the final subject and framing you want to inspect.
The camera [move and direction] at [pace], as [subject behavior]. The shot ends with [visible framing or revealed detail].
This is a planning aid, not a required syntax. Avoid adding a pan, orbit, zoom, and cut just to make a short clip sound cinematic. If a single move cannot communicate the idea, plan separate shots and assemble them in an editor.
3. Compare a fixed view with a small push-in
Imagine a reference photograph of a wooden chair beside a window, with a curtain and part of the room visible. The two original prompts below are unrendered planning exercises. They do not report tested model results. Keep the same reference, model, duration, and other settings when comparing them; keep the seed fixed if your service offers that control.
Brief A: keep the view fixed
The camera holds a fixed view. The chair stays in place while the curtain sways gently. The shot ends with the whole chair and window still visible.
Brief B: approach the chair
The camera makes a slow, small push-in toward the chair. The chair stays in place while the curtain sways gently. The shot ends with the whole chair and window still visible.
Only the camera instruction changes. That makes the comparison easier to interpret, although generated outputs may still differ in other ways. In A, check whether the room edges stay fixed. In B, check whether the view approaches the chair while preserving the requested ending. Neither result succeeds merely because something moves.
A push-in can crop the chair even if the opening photograph fits. Leave enough margin and choose a modest approach. Use the aspect-ratio and framing guide to prepare the reference for your final video shape.
4. Diagnose the mismatch before changing the prompt
- The subject moves instead of the camera: check both instructions. State the subject's intended behavior clearly.
- The move goes in the wrong direction: name its destination, not just “left” or “right.”
- The ending cuts off an important detail: reduce the move or prepare a reference with more surrounding room.
- A new angle changes the object: compare the silhouette and visible surfaces. Try a smaller arc if the revealed view is unsuitable.
- The clip changes unpredictably: remove competing actions, then change one instruction in the next attempt.
Watch the whole clip, then inspect the opening, middle, and ending. Save the prompt and settings with notes about what actually happened. For a product shot, the product-reference checklist also helps you check whether packaging and shape remain usable.
Draft the camera instruction in Shotmelo's image-to-video prompt planner, save it locally, and export the text plan for your chosen video service. Shotmelo's current preview supports planning and editing; it does not render videos or test whether a model follows a move.
Common camera questions
Is a pan the same as sideways tracking?
A pan rotates the view from one camera position. Sideways tracking moves the viewpoint along with the subject. Describe whether you want to look across a scene or travel beside someone.
Why does a dolly prompt look like a zoom?
A generator may approximate the request by enlarging the view. Look at how foreground and background details change relative to each other, rather than judging only the subject's size. If the distinction matters, record the mismatch and try a simpler approach with a reference that makes the depth easier to inspect.
Can a camera prompt guarantee a perfectly still shot?
No. Review the actual output even when you request a fixed camera. If small movement makes the clip unusable, assess another attempt or an editor's stabilization tools, including any crop they introduce.