AI AI Tools Guidebook
Prompt Library

Cinematic Camera Movement Prompts: 12 Director-Grade Moves

12 cinematic camera-movement prompts for Sora 2, Veo 3.1 and Runway Gen-4 — dolly, pan, tilt, crane, push, pull, rack-focus, whip-pan, Steadicam — each with explicit speed and duration.

Published: Updated: Author: AI Productivity Guide Team 🌐 查看中文版本

Camera movement is where AI video most often breaks. A vague “cinematic” prompt produces noodly motion that doesn’t read as any real camera. Directors use a small, precise vocabulary — dolly, pan, tilt, crane, push, pull, rack-focus, whip, Steadicam — because each term is unambiguous to a camera operator. The official Sora 2 prompting guide says the same thing in different words: give each shot one clear camera move and one clear subject action, and prefer a physical move (dolly, track) over an optical zoom. Naming the exact move and its speed is the difference between a deliberate shot and AI jelly. For broader fundamentals, see the AI video prompt basics walkthrough.

TL;DR

  • Name the specific move (dolly in, crane up, whip pan), not just “cinematic.”
  • Pair it with an explicit duration that fits your model’s clip limit (Sora 2: 4/8/12/16/20s; Veo 3.1: 4/6/8s; Runway Gen-4: 5/10s).
  • State one move per clip — stacking dolly + pan + tilt makes the model keep one and noodle the rest.
  • Say “dolly” (camera physically moves, parallax shifts) when you don’t want a “zoom” (lens crop, flat); the models treat them differently.
  • Add speed words (“slow,” “rapid,” “0.5-second push”) — slow 3-5s moves are noticeably more stable than fast ones.

Which model handles which move (June 2026)

Clip length is the hard ceiling on how much movement you can ask for. These are the current single-generation limits:

Model (June 2026)Single-clip lengthNative audioNotes for camera moves
Sora 2 / Sora 2 Pro4, 8, 12, 16, 20s (default 4s)YesStrong on parallax dolly + orbit; one move per shot
Google Veo 3.1 / 3.1 Lite4, 6, 8s (extend by chaining)YesSmooth slow moves; keep handheld subtle
Runway Gen-45 or 10sGen-4.5 onlyMotion Brush 3.0 to paint motion regions
Runway Gen-4.5up to 60s single genYesBest for sustained one-take moves

Pick a duration the model can actually hold. Asking Veo for an 8-second high-speed orbit, or Sora for a stable 20-second whip pan, pushes past the steady window and the motion degrades.

Best for

  • Cinematic shorts and music-video shots in Sora 2 or Veo 3.1
  • Establishing shots that need controlled reveal pacing
  • Mood pieces where movement does the storytelling
  • Product hero videos where one slow push communicates premium
  • Title sequences where the move itself is the punchline

1. Slow dolly in

cinematic video, slow dolly in toward subject, locked focus on subject's eyes, soft cinematic lighting, shallow depth of field, 4 seconds, static frame edges, no zoom

2. Slow crane up reveal

cinematic video, slow crane up from ground level revealing wide landscape, dawn light, atmospheric haze in distance, subject remains in lower third, 5 seconds, smooth motion

3. Rack focus pull

cinematic video, rack focus from foreground object (close, sharp) to subject in midground (initially soft), camera locked on tripod, soft window lighting, 3 seconds, no camera motion

4. Slow push in on face

cinematic close-up video, slow push in on subject's face, subject still, eyes catching key light, shallow depth, 50mm look, 4 seconds, no zoom — physical dolly only, ends just past head-and-shoulders framing

5. Lateral tracking shot

cinematic video, lateral dolly tracking subject walking left-to-right, parallax across background trees, 35mm lens look, golden hour, 6 seconds, constant speed, subject stays centered

6. Whip pan transition

cinematic video, whip pan left-to-right starting on subject A, blur through middle, ending on subject B in new location, motion blur during pan, 2 seconds total, designed for cut points at either end

7. Handheld documentary follow

handheld documentary video, operator following subject from behind through crowded market, gentle natural shake (not stylized), 35mm lens look, available light, 6 seconds, subject stays mid-frame

8. Steadicam orbit

cinematic video, Steadicam orbit 180 degrees around stationary subject, smooth gimbal motion, subject locked center frame, background revealing as camera moves, 6 seconds, no vertical bob

9. Tilt up reveal

cinematic video, slow tilt up from subject's hands working on craft to subject's face, camera body static, 50mm lens, soft directional light, 4 seconds, ends on eye contact

10. Pull out reveal

cinematic video, slow pull out (reverse dolly) from tight close-up to wide environment shot, subject reveals their context — workshop, street, room, 6 seconds, motivated reveal not arbitrary zoom

11. Locked-off long take

cinematic video, locked-off static frame, no camera movement whatsoever, subject performs action within frame (sits down, lights candle, looks up), composition does the work, 8 seconds, intentional stillness

12. Crash zoom (when justified)

cinematic video, crash zoom from wide to close on subject's reaction, rapid 0.5-second push, ends locked on face, motivated by sudden offscreen event, used sparingly, 3 seconds total clip

Dolly vs zoom: the one distinction that fixes most clips

The single most common reason a “cinematic” move looks fake is that the model rendered a zoom (the lens crops in, the image stays flat) when you wanted a dolly (the camera body moves through space, foreground and background slide past each other at different rates). The OpenAI cookbook’s fix is to describe the side effect, not the intent: write physical dolly forward, foreground objects passing on the sides, parallax effect rather than camera moves toward subject. Naming parallax is what tells the model to add depth instead of a flat crop.

Common mistakes

  • Stacking too many camera moves (dolly + pan + tilt) into one clip — the model picks one and noodles the rest.
  • Vague “cinematic” with no specific move named.
  • No duration — the model picks a generic length and pacing, usually too fast.
  • Asking for motion that exceeds the clip ceiling (a steady 20s whip pan, or an 8s high-speed orbit in Veo).
  • Confusing “zoom” with “dolly” — zoom is a lens crop, dolly is the camera physically moving; cue parallax to force the dolly.

FAQ

Do these prompts work the same in Sora 2, Veo 3.1 and Runway? The vocabulary transfers, but tune the duration. Sora 2 takes 4/8/12/16/20-second clips, Veo 3.1 caps a single clip at 8 seconds (chain to go longer), and Runway Gen-4 does 5 or 10 seconds. Set the number in the prompt to a length the model actually supports.

Why does my dolly look like a flat zoom? The model rendered an optical zoom instead of a physical move. Add the words parallax, objects passing on the sides, or foreground slides past background, and explicitly write no zoom — physical dolly only, exactly as in prompt 4.

How long can a single camera move stay smooth? Slow 3-5 second moves are the most reliable across all three tools. Fast moves (whip pans, crash zooms) hold best at 2-3 seconds. For anything longer than a tool’s native clip, use Runway Gen-4.5 (up to 60s) or chain Veo clips and manage continuity yourself.

Can I combine two camera moves in one shot? Prefer not to. Both directing practice and the Sora 2 guide recommend one camera move plus one subject action per shot. If you need a compound move (e.g. crane up into a push), shoot two clips and cut them together.

Should I specify a lens (35mm, 50mm)? Yes — a lens note nudges the field of view and depth of field. 35mm reads wider with more background; 50mm reads as a natural portrait look with shallower depth. It’s a cheap way to make the move feel intentional.

Tags: #Camera movement #Cinematic

Related Articles