Camera movement prompts for AI video: pan, dolly, orbit

A camera movement prompt names the move in plain film terms, one move per shot, with its direction and speed: “slow dolly in toward the mug”, “the camera orbits the car clockwise”, or “static shot, no camera movement”. With Sume's video generation API, the prompt is the only place to ask for a camera move, because the request has no camera-control setting. A first and a last frame can pin where the move starts and ends.

The request fields and prompt advice come from Sume's Video generation docs, read on 2026-09-28. The camera terms are general film language, and Sume's docs don't say how any model responds to them, so check each clip.

Is there a camera control setting?

No. None of the request parameters the docs list for POST /v1/videos moves the camera, and the fields are fixed: in current code, a field the API doesn't define is refused with a 400. The docs' prompt advice names the camera instead: include “details about motion, camera angles, lighting, and scene composition.”

What are the main camera movement terms?

Write the term with a direction and a speed. The wording column shows one way to phrase each move:

How do I write a camera movement prompt?

Describe the shot first, then the move. Two example requests in Sume's docs put the move in words: “Gentle camera drift; keep the product locked in frame” and “Slow push-in on a ceramic mug”. Both are written for the retiring Video 1.0 route; the same kind of wording goes in the prompt of a POST /v1/videos request:

• One move per shot, with its direction and speed.
• Say what the move is relative to: the subject, and where it ends up in the frame.
• Say what must hold still while the camera moves, as the docs' product example does.
• Fit the move to the clip's length: one clip runs 2–30 seconds, depending on the model.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: orbit-shot-001" \
  -d '{
    "model": "seedance-2",
    "prompt": "A red vintage car parked on an empty desert road at sunset. The camera slowly orbits the car clockwise at waist height.",
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "duration": 8
  }'

How do I control where the move starts and ends?

Give the model both ends. Send the opening framing as the first_frame and the closing framing as the last_frame, then describe the move between them: a wide shot and a close-up of the same scene set up a push-in. Today every video model except grok-imagine-video-1.5 takes a last frame, and in current code a last frame needs a first frame. Image-to-video API: first and last frame has the fields, and Image to video prompt examples has more wording.

What about a dolly zoom or a drone shot?

A dolly zoom moves the camera toward or away from the subject while the lens zooms the other way, so the subject stays about the same size and the background seems to stretch or shrink. Ask for it the same way, as one move, and check the clip. Aerial moves and zoom-outs built from a photo have their own posts: AI drone shot from a photo and Earth zoom out AI video.

Plan the move when you generate the clip, not when you assemble the edit: in a Timeline 1.0 render, a still is a static hold, and its motion field is accepted and ignored.