Cinematic Camera Movement AI Video Prompts: Dolly, Tracking, Orbit, and More
TL;DR
- Camera direction is the highest-leverage prompt element: leave it out and models default to a static, drifting mid-shot — a slideshow, not cinema.
- Control runs on three axes — movement, framing (CU/MS/WS), and lens (35mm, 85mm, macro) — each a separate clause.
- Sixteen core moves and twelve camera-family templates cover dolly, tracking, orbit, crane, whip pan, FPV fly-through, macro glide, and static lock-off — with style modifiers (anamorphic, golden hour, shallow DOF) completing the cinematic look.
- Parameters matter as much as the move name: speed words, start/end cues (“begins wide, ends on her eyes”), and continuity with subject motion (parallel vs. contrast).
- Most failures come from ambiguous wording — rack focus vs. dolly, zoom vs. push-in, FPV vs. drone — each with a fix.
Why camera direction is the highest-leverage prompt element
Write a prompt with no camera instruction and something predictable happens: a static mid-shot with a slow, directionless drift. Nothing moves with intention — technically video, cinematically nothing.
The model defaults to the safest interpretation: “a woman walking through a market” says nothing about the camera, so the camera parks at eye level. Google’s Veo cookbook treats camera movement as first-class prompt notation — pan, tilt, dolly, and lens specs — because the camera is a character in the prompt, not an afterthought Veo cookbook.
Camera direction is the highest-leverage element for three reasons:
- It determines perceived production value. A locked-off phone shot and a slow dolly-in feel like different budgets; movement is the cheapest signal a crew made this.
- It controls emotional trajectory. Framing says what to look at; movement says how to feel over time. A push-in escalates intimacy; a pull-back isolates.
- It anchors subject motion. A camera that moves with intention gives the model a scaffold for rendering the subject coherently; a drifting camera produces blurred limbs and slipping geometry.
This guide to cinematic camera movement AI video prompts — dolly, tracking, orbit, and more — is a working system: shared vocabulary, accepted parameters, misread wordings, and twelve ready templates. Annotated cinematography guides treat aperture, focal length, and move as separate specs; the same layering applies here Google AI Studio Veo prompt cookbook.
The camera movement vocabulary
Each row gives a definition, exact prompt phrase, emotional effect, and risk level — how likely models misrender the move when you name it bare. “High” means it needs speed words or start/end cues, not that it is unusable.
| # | Move | Definition | Prompt phrase | Emotional effect | Risk level |
|---|---|---|---|---|---|
| 1 | Dolly in | Camera moves toward the subject on an axis | “slow dolly in on the subject” | Intimacy, rising tension | Low |
| 2 | Dolly out | Camera retreats from the subject | “gradual dolly out revealing the room” | Reveal, isolation, context | Low |
| 3 | Tracking (lateral) | Camera moves sideways parallel to the subject | “lateral tracking shot moving with the runner” | Momentum, journey | Medium — often substituted by a pan |
| 4 | Orbit / arc | Camera circles the subject at constant radius | “slow 180-degree orbit around the statue” | Heroism, 3D reveal | Medium — arc collapses into a pan |
| 5 | Crane / jib | Camera rises or descends on a vertical arc | “crane up from the crowd to a wide of the skyline” | Epic scale, reveal | Medium — needs start and end frame |
| 6 | Pedestal | Camera moves straight up/down without tilting | “pedestal up to eye level” | Reveal through elevation | Medium — confused with tilt |
| 7 | Pan | Camera rotates horizontally on a fixed axis | “slow pan left across the empty street” | Surveying, anticipation | Low |
| 8 | Tilt | Camera rotates vertically on a fixed axis | “tilt up from her hands to her face” | Anticipation, scale | Low |
| 9 | Handheld | Camera carries organic operator shake | “handheld with subtle operator shake” | Urgency, realism | Medium — models over-shake |
| 10 | Static / lock-off | Camera fixed; motion comes only from subject | “locked-off static shot, tripod-mounted” | Objectivity, patience | Low |
| 11 | Whip pan | Very fast pan with motion blur between subjects | “whip pan from the driver to the passenger” | Energy, kinetic transition | High — blur hard to control |
| 12 | Push-in | Gradual tightening of frame toward the subject | “slow push-in to a close-up” | Emotional escalation | Low |
| 13 | Pull-back | Gradual widening away from the subject | “pull-back to reveal the full cathedral” | Consequence, perspective | Low |
| 14 | Zoom | Focal length changes while camera stays fixed | “slow zoom in on the photograph” | Unease, surveillance | High — models default to dolly |
| 15 | FPV fly-through | First-person drone view racing through space | “FPV drone fly-through threading the arches” | Adrenaline, immersion | High — geometry collisions common |
| 16 | Macro glide | Very close, very slow move across a small surface | “macro glide across the watch face” | Tactile luxury, precision | Medium — focus plane slips |
Three notes on the vocabulary:
- Dolly vs. zoom vs. push-in is the table’s key distinction. A dolly moves the camera through space (perspective shifts); a zoom changes focal length from a fixed position (perspective compresses); a push-in is the intent, achievable by either. Conflating them produces inconsistent output — product-video libraries name the mechanism, not the outcome Google AI Studio Veo prompt cookbook.
- Orbit is the workhorse hero move — always add a degree count (“90-degree,” “180-degree”); an open-ended “orbit” invites the model to spin past your clip length.
- Static is a choice, not an absence. Naming it explicitly suppresses default drift and forces all motion onto the subject.
Risk levels are editorial judgments from repeated prompting, not benchmark scores.
Two-axis control: movement + framing
Beginners give one camera instruction. Professionals give three, on separate axes:
| Axis | Question it answers | Vocabulary | Example phrase |
|---|---|---|---|
| Movement | What does the camera do over time? | dolly, track, orbit, pan, tilt, crane, static | “slow dolly in” |
| Framing | Where does the subject sit in frame? | CU (close-up), MS (medium shot), WS (wide shot), OTS | “ending in a close-up” |
| Lens | How does the glass render space? | 35mm, 50mm, 85mm, macro, anamorphic, telephoto | “85mm lens, shallow depth of field” |
Sentence pattern: [speed] + [movement] + [direction/target] + [framing arc] + [lens/look]. Example: “Slow dolly in from a wide shot to a close-up on her eyes, 85mm lens, shallow depth of field.”
Separating the axes matters because movement and framing are constantly confused. “The camera moves closer” is movement; “the subject is framed in close-up” is framing — and a model can satisfy one without the other (zooming to fake a dolly, or cropping rather than moving). Lens governs what movement looks like: a dolly on a 24mm feels immersive; on an 85mm it feels compressed and portrait-like.
Aspect ratio note. Framing words depend on aspect: a “wide shot” in 16:9 breathes on both sides; in 9:16 it can imply full-body framing. For vertical targets (Shorts, Reels, TikTok), state “vertical 9:16 framing” before choosing shot sizes, or the model may compose for landscape and crop. In multi-shot work, each shot needs its own movement/framing/lens reset, not one global clause — the discipline our multi-shot storyboard article treats as core to per-shot camera control.
Movement parameters that models accept
Speed words
Speed is the most reliable movement parameter:
| Speed word | Feel | Use when |
|---|---|---|
| “static,” “frozen” | No movement at all | You want zero camera drift |
| “slow,” “gentle,” “subtle” | Cinematic ease across the full clip | Drama, beauty, product, establishing |
| “gradual,” “steady” | Deliberate, linear progression | Reveals, crane shots, walk-and-talks |
| “brisk,” “smooth” | Confident but not urgent | Commercial, travel, lifestyle |
| “rapid,” “fast,” “aggressive” | Kinetic, high-energy | Action, whip pans, FPV |
Avoid unquantified extremes: “very fast dolly” invites warping; “the slowest possible orbit” often stalls. When a move must finish within a short clip, say so: “completes a full 90-degree orbit within the shot.”
Start and end cues
Anchoring the first and last frame is your strongest control over long moves: “begins …, ends …” or “from … to …”.
- “Begins on a wide of the plaza, ends in a close-up on the ring in her hand.”
- “Starts at street level, cranes up and ends level with the penthouse balcony.”
They also solve truncation: a model that knows the landing point paces the middle instead of running out of clip halfway through.
Continuity with subject motion
The subject is moving too. Two deliberate relationships:
- Parallel (matching): camera and subject move the same direction at similar speed — “lateral tracking shot matching her walking pace.” The audience rides with the subject: the walk-and-talk, the chase, the promenade.
- Contrast (opposing or still): camera moves against the subject, or holds — “static locked-off shot as she walks away into the fog.” Effect: tension, alienation, formal composition.
Contrast is higher-risk — it needs two independent motion fields. If a take collapses, simplify to parallel or state it outright: “camera moves opposite to the subject’s direction.” When reverse-engineering movement from footage, this lens is how to classify what you see — our reference-video-to-ai-prompt conversion guide builds a full prompt from that extraction.
Movement prompts that models misread
Four confusions account for most camera-direction failures. Each has a wording fix.
| You write | Model often hears | What you get | Precise rewording |
|---|---|---|---|
| “rack focus to the door” | A dolly toward the door | Camera moves; focus never shifts | “camera stays fixed; focus racks from the foreground cup to the door” |
| “zoom in on her face” | A push-in / dolly | Perspective expands instead of compressing | True zoom: “optical zoom from a fixed position, perspective compresses.” Push-in: name it a push-in, drop “zoom” |
| “FPV drone shot flying through” | A conventional aerial shot | Slow, high-altitude survey footage | “FPV first-person view, fast low-altitude flight threading between the pillars, motion blur at the edges” |
| “the camera follows her” | Unspecified follow — often a pivot | Subject centered while background pivots | “tracking shot moving alongside her” (parallel) or “camera pans to follow her as she passes” (pivot) |
Two broader habits:
- State what does NOT move when ambiguity is a risk. “Locked-off camera, no movement” is load-bearing, not redundant — negative instructions suppress default drift. Guidance for current models agrees: unspecified behavior is filled in with the model’s default, so every unstated axis is a gamble MStudio, how to prompt AI video models.
- One move per shot clause. “Dollies in, then orbits, then whips left” renders as one confused half-move. Choreography belongs in multi-shot sequences — one move per shot.
12 prompt templates by camera family
Substitute the subjects and environments; keep the camera clauses intact.
Dolly family
1. Push-in reveal (escalating intimacy)
Template: Slow dolly push-in from a medium shot to a tight close-up on the woman's eyes as she looks up from the letter, 85mm lens, shallow depth of field, golden hour window light from camera-left, one continuous move, nothing else.
2. Dolly-out context (isolation reveal)
Template: Gradual dolly out — begins close on the musician's hands on the guitar strings, ends in a wide shot revealing the empty concert hall around him, 35mm lens, warm practical lights, steady linear speed, camera moves straight back on axis with no pan or tilt.
3. Dolly-zoom vertigo (disorientation)
Template: Hitchcock-style dolly zoom on the detective in the corridor — camera physically moves forward while the lens zooms out simultaneously, so his face stays the same size while the hallway behind him stretches and distorts, 50mm lens, slow unsettling pace, framing locked on his face throughout.
Tracking family
4. Lateral walk-and-talk (parallel motion)
Template: Lateral tracking shot moving parallel with two friends walking along the waterfront, matching their pace exactly, medium two-shot framing held constant, 35mm lens, late afternoon sun flaring from behind them, smooth Steadicam feel, background streams past horizontally, no zoom.
5. Follow-cam (behind-the-subject pursuit)
Template: Follow-cam tracking shot from behind the runner as she sprints through the rainy neon alley, camera moving forward at her speed, over-the-shoulder wide framing, subtle handheld sway, ends as she reaches the glowing intersection.
6. Hand tracking product (commercial precision)
Template: Smooth tracking shot gliding left-to-right past three sneakers on concrete blocks, camera on a straight horizontal line at constant slow speed, each shoe fills the frame briefly at medium close-up, studio softbox light, 50mm lens, shallow depth of field, ends on the third sneaker.
Orbit / crane family
7. Hero orbit (product or character)
Template: Slow 180-degree orbit around the chef at the kitchen pass, camera circles at waist height keeping him centered in a medium shot, completes the half-circle within the shot, background kitchen lights streaking softly behind him, 35mm lens, warm tungsten ambience, constant radius, no zoom, no crane.
8. Crane reveal (scale shift)
Template: Crane up from a low close-up of the bride's hands on the bouquet, rising steadily to a high wide shot revealing the cliffside ceremony and ocean beyond, begins tight ends wide, gradual speed, golden hour light, camera rises vertically and tilts down slightly as it clears the crowd, anamorphic look.
9. FPV flythrough (architectural adrenaline)
Template: FPV drone first-person view racing through the abandoned factory — threads the broken windows, banks left around the central column, skims the flooded floor, fast low-altitude flight with motion blur at the frame edges, wide 16mm lens, dust beams in shafts of light, ending as it bursts out through the roof opening into daylight.
Static + subtle family
10. Locked-off with subject motion (contrast)
Template: Locked-off static shot, tripod-mounted, no camera movement — medium-wide framing of the empty train platform as the woman walks from the far right of frame to the far left and exits, 35mm lens, overcast diffused light, all motion comes from the subject.
11. Slow tilt (anticipation)
Template: Slow tilt up from the muddy boots on the forest floor, traveling up the rain-soaked coat to settle on the climber's face, locked vertical axis with no panning and no dolly, begins at ground level ends on her eyes, gradual deliberate speed, cool desaturated grade, 50mm lens.
12. Macro glide (tactile detail)
Template: Macro glide moving slowly across the face of the mechanical watch — camera travels in a shallow arc over the dial and gears, extreme close-up with razor-thin depth of field, focus holding on the moving balance wheel, soft raking light catching the brushed metal, 100mm macro lens, unhurried pace, no zoom, no tilt.
Camera + style combinations that sell the cinematic look
Movement is necessary but not sufficient: use one movement clause + one style clause + one lens clause, or prompts start contradicting each other.
| Style modifier | Prompt add-on | Pairs best with | Effect on the move |
|---|---|---|---|
| Anamorphic | “anamorphic 2.39:1 widescreen, horizontal blue lens flares, oval bokeh” | Dolly, crane, pull-back | Widening moves gain scale; flares streak as lights cross frame |
| Golden hour | “golden hour side light, long soft shadows, warm amber grade” | Tracking, orbit, push-in | Moving camera constantly re-renders the light angle |
| Shallow DOF | “shallow depth of field, creamy bokeh, focus locked on the subject” | Dolly in, macro glide, tracking | Movement shows through shifting defocus — the “expensive” signal |
| Handheld documentary | “handheld with organic breathing, subtle micro-shakes” | Walk-and-talk, follow-cam | Adds immediacy; cap with “subtle” to avoid shake overload |
| 35mm film grain | “35mm film emulation, fine grain, gentle halation” | Static lock-off, slow pan | Textures still frames so slow moves don’t read as sterile CGI |
| Neon night practicals | “night city, neon practicals, wet reflections, teal-magenta grade” | FPV, whip pan, orbit | Light streaks make speed legible; blur resolves into color |
Two signature combinations: shallow-DOF dolly in at golden hour reads instantly as high-production, and anamorphic crane reveal equals epic scale — but state its start and end frames or the crane won’t complete its arc. Two practical notes: anamorphic loses its identity cropped to 9:16 (swap to “shallow DOF + film halation” for vertical), and with native-audio models, movement should match sound design — a whip pan wants a whoosh, a slow dolly swelling room tone (more on native audio).
FAQ
1. Why does my AI video look static even when the scene is interesting? Because you did not name the camera: unspecified behavior defaults to a static or drifting mid-shot. The fix is one explicit movement clause with a speed word — “slow dolly in, 35mm” — before anything else.
2. What is the difference between a dolly-in, a push-in, and a zoom? A dolly-in physically moves the camera toward the subject, so background perspective shifts. A zoom changes focal length from a fixed position, so the background compresses instead. A push-in is the general intent of tightening on the subject, achievable either way — so name the mechanism: “camera moves forward on axis” vs. “optical zoom from a fixed position.”
3. How many camera moves can I put in one prompt? One move per shot clause. “Dolly in, then orbit, then whip pan” renders as one confused gesture in a short clip. Reserve choreography for multi-shot sequences with a camera reset per shot, or for one move with staged beats (“begins wide, ends on her eyes”).
4. Which camera moves are riskiest in current models? Whip pan (blur control), true zoom (models substitute a dolly), open-ended FPV fly-through (geometry collisions), and pedestal (routed to a tilt). Same mitigation for all four: speed words, start/end cues, and a negative clause.
5. Do camera prompts differ between Veo, Kling, Seedance, and Wan? The core vocabulary transfers; tolerance for detail differs. Models with documented camera notation (Veo’s cookbook uses pan/tilt/dolly plus lens specs) accept dense cinematography; others prefer one or two clauses. Start shared, add layers only if the model honors them.
6. How do I keep camera movement consistent across a multi-shot video, and does wording change for natural-language formats? Give every shot its own movement/framing/lens clause instead of one global instruction, and reuse identical wording for moves you want repeated (e.g., “slow 90-degree orbit” in every product shot). Extracting moves from footage? See the reference-to-prompt workflow and the multi-shot storyboard method. Natural-language prompts (our 2026 natural prompt formula) still need the same camera verbs — you describe the shot the way a director briefs a DP.
Conclusion
Cinematic camera movement in AI video prompts comes down to three disciplines: speak the vocabulary — name the move, since unspecified cameras default to static mid-shots; parameterize it — speed, start/end cues, a defined relationship to subject motion; disambiguate it — separate movement from framing from lens, and refuse compound choreography in single shots.
The 16-move table is your reference, the 12 templates your starting blocks, and the style stack what turns a correct move into a shot worth posting. Practice on one subject with three moves from three families; the emotional difference between a dolly-out, an orbit, and a lock-off teaches more than theory.
Keep reading: - AI video prompts: the natural 2026 formula — camera direction in conversational prompt structure. - Multi-shot storyboard AI video prompts — resetting the camera per shot. - Veo 3.1 native audio + 4K — pairing camera movement with sound. - Hotel & resort promo video prompts — drone camera verbs for property tours. - Reference video to AI prompt conversion — extracting camera motion from references.
Reviewed by the videosprompt.org editorial team · October 2026
Share Article
