Camera Movement Video Prompts: Pan, Dolly, Orbit Keywords
Camera movement prompts for Kling, Runway, Luma, and Pika: pan, dolly, and orbit keywords that lock one path without fighting subject motion.
Turn any photo into an AI prompt — free
No sign-up required. Works with Midjourney, FLUX, DALL-E.
Try Image to Prompt →Camera movement prompts tell video hosts how the lens travels: pan, dolly, orbit, truck, tilt, or locked tripod. Kling AI, Runway Gen-4, Luma Dream Machine, Pika, and Hailuo treat "cinematic camera" as random drift. Name one path, one direction, and one speed matched to the UI seconds. This guide is a pan, dolly, and orbit keyword reference for video hosts as of August 2026. You leave with paste-ready movement lines, stack order, and model dialect notes. Soft path when the clip starts from a hero still: PromptMake /image. Still framing belongs on the camera-angle guide. Aspect, grain, and lens look belong on the cinematic video keywords guide.
What camera movement prompts do for video hosts
Camera movement prompts describe how the virtual camera changes position or angle across frames. Subject motion is what the actor or prop does. Camera movement is what the lens does. "Runner walks toward camera" is subject motion. "Slow dolly in only" is camera movement. Video models merge the two when you write "dynamic cinematic reveal," so the prop spins while the lens orbits and both fail.
Searchers who type camera movement prompts want named moves that survive a first credit spend. Forum strings often stack pan plus dolly plus orbit in one five-second box. Short hosts punish that stack. One path per generation pass is the rule that travels across Kling, Runway, Luma, Pika, and Hailuo as of mid-2026.
Strong fit: creators cutting hero clips, marketers testing product spins, designers animating Midjourney or FLUX stills, editors who need a clean push-in before a CapCut cut. Skim if you only export stills. Skip this page if you only need viewpoint language for Midjourney stills. That job is height, lens, and shot size on a frozen frame.
Write movement after subject nouns and timed action beats. Write look keywords after movement. Duration lives in the host slider and in your beat list. Movement keywords do not replace duration. They spend the seconds you already chose.
Pan and tilt keywords
Pan rotates the camera left or right on a fixed stand. Tilt rotates the camera up or down on that same stand. Neither move requires the camera body to travel through space. That fact matters on short video hosts. Pan and tilt cost less spatial invention than a full orbit, so they hold subject identity better on image-to-video uploads.
Weak pan language: "sweeping cinematic camera." Strong pan language: "slow pan left only," "gentle pan right across the shelf," "static camera then pan left from two to five seconds." Add direction, speed, and a timing window when the host supports multi-second clips. Speed words that travel: slow, gentle, measured. Speed words that break short clips: whip, crash, frantic.
Pan works when the subject stays mostly fixed and the environment reveals. A product on a table, a skyline, a mural wall, a row of bottles. Pan fails when you also demand a walking subject that exits frame. Pick reveal-by-pan or walk-by-subject. Do not ask for both in three seconds.
Horizontal pan left and right
Pan left keywords: slow pan left only, gentle pan left across scene, camera pans left on tripod, horizontal pan left, locked height. Pan right keywords: slow pan right only, gentle pan right, tripod pan right, no tilt. Paste example for Kling text-to-video: "Medium shot of a wet city street at dusk, yellow taxi parked curb-side, slow pan left only across storefront neon, 5 seconds."
Name the start framing when the pan reveals something off-screen: "Start on the mug, slow pan right to the window over three seconds." That start-end language helps Runway Gen-4 and Luma hold the first subject before the lens turns. Without a start anchor, the model may begin already mid-pan.
Vertical social lanes (9:16) still accept horizontal pan, but the travel distance shrinks. Keep pan arcs short on vertical clips. A long pan left on 9:16 exposes empty side walls the model invents. Prefer a short pan that stays on subject mass.
Whip pan keywords (whip pan left, crash pan) belong on stylized music or sports cuts with intentional blur. Skip whip language on product label reads and portrait image-to-video. Blur eats type and face detail.
Tilt up, tilt down, and Dutch caution
Tilt up keywords: slow tilt up only, gentle tilt from ground to sky, camera tilts up on locked tripod. Tilt down keywords: slow tilt down only, gentle tilt from skyline to street, tilt down to product on table. Paste example for Luma: "Locked tripod under a glass tower facade, slow tilt up from lobby doors to glass crown over six seconds, subject architecture only."
Pair tilt with a motivated reveal. Tilt up on a building sells height. Tilt down on a tabletop sells the product. Unmotivated tilt ("dramatic tilt") returns random pitch with no story beat.
Dutch tilt (canted horizon) is a spice word for unease. Keywords: slight Dutch tilt, canted frame. Leave Dutch tilt out of catalog, headshot, and ecommerce clips. Most short hosts treat Dutch language as a continuous roll that fights readable labels.
Do not combine pan and tilt in one five-second pass unless the brief needs a diagonal reveal and you have tested that host. "Slow pan left while tilting up" is two axes. Split into two clips or extend duration past eight seconds on Runway Gen-4 before you stack axes.
Dolly, truck, and push-in keywords
Dolly moves the camera body through space toward or away from the subject. Truck (also called tracking left or right) moves the camera body sideways. Push-in and pull-out are everyday synonyms for dolly in and dolly out. These moves change perspective. Near objects grow or shrink. Background parallax appears. That parallax is why dolly reads more "cinematic" than a simple digital zoom crop.
Video hosts confuse dolly with zoom when you write "zoom in cinematic." Prefer dolly or push-in language when you want parallax. Prefer "locked camera, optical zoom feel" only when you accept flat crop behavior. Most consumer tiers fake zoom as a crop. Dolly keywords push the model toward spatial travel.
Dolly works on product heroes, portrait approach, hallway walks, and tabletop reveals. Dolly fights fast subject motion. A sprinter plus dolly in on a three-second Hailuo clip often smears legs. Lock the subject or slow the subject when dolly is the star move.
Dolly in and dolly out
Dolly in keywords: slow dolly in only, gentle push-in, camera dollies toward subject, slow push-in from medium to close. Dolly out keywords: slow dolly out only, gentle pull-back, camera dollies away from subject, pull-out revealing room. Paste example for Runway Gen-4: "Medium shot of brushed steel bottle on gray stone, bottle static, slow dolly in only from medium to close over five seconds, soft key from upper left."
Write start and end shot size when the host supports longer clips: "Start medium, end close-up, slow dolly in only." That pair stops the model from beginning already tight. Image-to-video loves this pattern: preserve the upload, subject static, slow dolly in only.
Speed discipline: slow and gentle travel. "Aggressive dolly slam" on a face upload destroys identity on Kling and Luma. Product labels need slow push-ins so type stays readable through the move.
Crane and boom language (crane up, boom down) lifts the camera on a vertical path through space. Use on exteriors and architecture when duration is six seconds or more. Skip crane stacks on five-second Pika product loops. Pika short lanes reward push-in or locked camera more than crane ballet.
Lateral truck and tracking shots
Truck left keywords: slow truck left only, lateral track left, camera trucks left parallel to subject. Truck right keywords: slow truck right only, lateral track right, sideways tracking shot right. Paste example for Kling: "Person in olive coat walks right on wet crosswalk; camera trucks right parallel to walker, matching pace, five seconds."
Matching truck to a walking subject is a classic tracking shot. Write pace language: "camera trucks right matching subject walk speed." Without pace language, the camera may lead or lag and crop the head.
Truck past a static product reveals context: "Bottle static on table; slow truck left revealing window light." That pattern keeps the hero sharp while the environment changes. It beats orbit when face or logo identity must hold.
Avoid truck plus dolly in the same pass on short clips. Lateral plus forward travel is a compound move. Hosts split the difference into diagonal mush. Pick truck or dolly. Combine in edit across two generations if you need both.
Orbit and arc camera keywords
Orbit circles the camera around a subject while keeping the subject centered. Arc is a partial orbit: a quarter turn or half turn instead of a full loop. Orbit sells dimensionality on products, cars, sculptures, and fashion turns. Orbit is also the move that destroys face identity most on image-to-video when you ask for a full 360 in five seconds.
Weak orbit language: "epic 360 cinematic orbit crane swoop." Strong orbit language: "slow quarter-orbit right around product only," "gentle arc left 90 degrees," "partial orbit clockwise, subject centered, five seconds." Name angle of travel. Quarter and half orbits land. Full 360 on short free tiers flickers or morphs materials.
Orbit wants a clear center subject and a simple background. Busy crowds behind an orbiting camera invite warping bystanders. Studio void, seamless paper, or soft bokeh plates hold better.
Full orbit vs partial arc
Full orbit keywords: slow 360 orbit clockwise, full orbit around subject, camera circles subject once. Use full orbit on products and objects when duration is eight to ten seconds on Runway Gen-4 or paid Kling lanes. Paste example: "Matte ceramic vase on seamless gray, slow 360 orbit clockwise, vase static, even studio softbox, 10 seconds."
Partial arc keywords: slow quarter-orbit right, gentle 90-degree arc left, partial orbit clockwise stopping at three-quarter view, arc from front to profile. Partial arcs fit five-second Pika and Luma clips. They show form without demanding a seamless loop closure the model often fails.
Clockwise vs counter-clockwise: pick one word and stick to it across retries. "Orbit around" without direction returns random spin. Fashion and car clips often prefer a three-quarter stop: start front, arc to three-quarter, hold last frame for thumbnail.
Handheld orbit micro-moves (subtle handheld orbit, slight arc with micro-shake) add documentary feel. Keep shake subtle. Heavy handheld plus orbit doubles motion blur on faces.
When orbit fights face identity
Portrait image-to-video with full orbit is a common credit sink. As the camera rounds the head, models invent ear shape, hair volume, and jawline. Fix: partial arc of 30 to 90 degrees, or lock the camera and add micro subject motion (blink, breath, smile growth) instead.
Preserve language must lead when you upload a face: "Maintain face, hair, and wardrobe from reference. Slow quarter-orbit right only. Subject mostly still." Drop orbit before you drop preserve if identity slips.
Product logos on orbit need blank or simple print sides. Busy typography wrapping a bottle will swim during orbit. Prefer dolly in on a fixed label face when type must stay sharp.
Orbit plus subject walk is two motion systems. "Person walks while camera orbits" collapses on Hailuo and Pika short lanes. Orbit a static pose, or track a walker with truck. Separate the jobs.
How to stack camera movement with subject and duration
Stack order keeps retries sane. Subject nouns first. Timed subject motion second. One camera movement third. Aspect and lens fourth if you need look control. Grain last. Duration in the UI and echoed once in prose. This order mirrors how Kling, Runway, Luma, and Pika parse public examples as of mid-2026.
Example stack for a six-second Runway street clip: "Low wide shot of wet asphalt crosswalk at dusk, yellow taxi static for two seconds then brake lights brighten, slow dolly in only, 16:9 frame, 24mm feel, fine film grain." Subject and light event come before the dolly clause. The dolly has one path and one speed.
Cut movement complexity before you cut subject detail when a render fails. If the clip drifted, lock the camera and retest subject motion alone. If the subject held but the frame felt dead, add one slow push-in on retry two. Add pan or arc only after push-in is clean.
Step-by-step paste order
- Write subject noun, material, count, and place.
- Write timed subject motion beats matched to UI seconds.
- Pick one camera path: locked, pan left, tilt up, dolly in, truck right, quarter-orbit.
- Add direction and speed: slow, gentle, matching walk pace.
- Echo duration once: "Five seconds total."
- Add aspect and lens only if the look layer matters this pass.
- Read aloud. Delete a second camera verb if two appear.
Merge into prose for hosts that reject labels: "Medium studio shot of a brushed steel bottle on gray stone. Bottle static. Slow dolly in only. Soft key from upper left. Five seconds." Keep the same order every time so edits stay predictable.
Image-to-video vs text-to-video
Text-to-video invents every pixel. Camera movement must be explicit because no upload locks composition. Name framing plus move: "medium shot, slow pan left only."
Image-to-video inherits framing from the still. Lead with preserve language. Add one camera path or "camera locked." Micro subject motion plus locked camera often beats orbit for faces. Crop the still to output aspect before upload so a pan does not invent empty walls.
Hero still quality gates the move. Soft edges and busy backgrounds break during dolly and orbit. PromptMake /image drafts Midjourney, FLUX, DALL·E, Stable Diffusion, or Leonardo prompts from a reference when you need a cleaner hero before Kling or Luma. Soft start: https://promptmake.net/image. Recreate Exactly holds layout. Adjust Lighting fixes exposure that would smear under a push-in. Create Variation explores alternate stills that share one movement shell.
Model notes for Kling, Runway, Luma, and Pika (2026)
Public model names and credit tables shift. Treat the notes below as an August 2026 snapshot. Confirm duration pickers and camera controls inside each host before you paste month-old forum strings.
Shared rule across hosts: one camera path per generation. Pan, dolly, and orbit are not interchangeable synonyms for "make it move." Each path spends spatial budget differently. Free tiers favor five-second tests. Prove the move on a short clip before you buy a ten-second orbit.
Hailuo joins the same grammar on short consumer lanes. Prefer locked or slow push-in on three-to-five-second Hailuo tests. Save full orbits for longer paid tiers when you have confirmed the host holds materials.
Kling and Runway
Kling AI reads clause-separated movement well. Put the move in its own short phrase: "Slow dolly in only." Follow with subject motion in a separate clause. Kling image-to-video holds upload identity better when orbit stays partial and speed stays slow.
Runway Gen-4 targets film vocabulary. Dolly, tracking, crane, and arc language align with Gen-4 training examples in mid-2026 docs. Runway tolerates slightly longer clips (eight to ten seconds) for a slow push-in with a single light event. Still keep one path. Gen-4 does not rescue a pan-plus-orbit stack.
Both Kling and Runway expose duration in UI. Echo seconds in prose on first tests so beats and slider match. Log which host ignores prose duration so you know when the slider alone must carry time.
Luma, Pika, and Hailuo
Luma Dream Machine favors readable sentences on five-second clips. Lead with subject and motion in plain prose. Trail the camera clause: "...slow pan left only, camera height locked." Luma image-to-video works best with subtle moves. Heavy 360 orbit on a portrait upload fights reference pixels.
Pika focuses on short social clips as of August 2026. Keep movement stacks short: one verb, one direction, explicit seconds. Pika duration limits reward push-in, locked tripod, or a short pan. Save full orbit language for hosts with longer lanes unless a paid Pikaframes route gives you room.
Hailuo short tests punish compound moves. Start with locked camera plus one subject beat. Add slow dolly in on pass two. Add orbit only after dolly is clean. Port the same movement line across hosts by shortening travel angle for Pika and Luma and allowing longer arcs on Runway.
Common camera movement prompt mistakes
Mistake 1: "Cinematic camera" with no path. Name pan, dolly, orbit, truck, tilt, or locked.
Mistake 2: Three moves in one five-second clip. Orbit plus dolly plus tilt collapses. One path per pass.
Mistake 3: Mixing subject walk with orbit. Pick tracking truck or static orbit.
Mistake 4: Full 360 orbit on a face upload. Use quarter-arc or locked camera with micro motion.
Mistake 5: Whip pan on product labels. Slow moves keep type readable.
Mistake 6: Zoom language when you wanted parallax. Prefer dolly or push-in.
Mistake 7: Movement before subject nouns. Anchor the object, then move the lens.
Mistake 8: Duration longer than the move needs. Ten seconds of UI time with a one-second pan wastes a frozen tail. Match seconds to travel.
Mistake 9: Aspect mismatch on pan. Crop the still to 9:16 or 16:9 before a reveal pan so the host does not invent walls.
When PromptMake /image fits before motion
Video hosts animate pixels. They do not fix a soft hero still. PromptMake /image helps you lock a clean frame before you open Kling, Runway, Luma, or Pika.
Use /image when you need a Midjourney or FLUX still with clear edges for a dolly, when a phone photo is too noisy for orbit, or when you want formatted prompt text from a reference without hand-translating model dialects.
Workflow: upload or build a still, copy formatted prompt text, iterate until the subject reads at thumbnail size, export PNG cropped to target aspect, upload to image-to-video, paste one camera movement line plus timed subject beats.
Goal modes map to movement prep. Recreate Exactly holds layout for uploads you plan to dolly into. Create Variation explores heroes that share one pan or arc shell. Adjust Lighting fixes blown highlights before a push-in magnifies them.
Guest accounts receive about three /image runs per day. Free registration raises image quota separately from /text. PromptMake does not render video. It shortens the still step so camera movement prompts attach to a frame worth animating. Soft start: https://promptmake.net/image.
FAQ
What are camera movement prompts?
Camera movement prompts are text lines that tell an AI video host how the lens travels: pan, tilt, dolly, truck, orbit, arc, or locked tripod. They sit beside subject nouns and timed action beats. Strong lines name one path, one direction, and one speed matched to the clip duration.
How do pan prompts differ from dolly prompts?
Pan rotates the camera on a fixed stand left or right. Dolly moves the camera body through space toward or away from the subject. Pan reveals a scene without parallax. Dolly changes perspective and adds parallax. Pick pan for shelf or skyline reveals. Pick dolly for product push-ins and portrait approach.
When should I use orbit camera keywords?
Use orbit or arc when you need to show three-dimensional form on a product, vehicle, or sculpture. Prefer partial arcs (90 degrees or less) on five-second clips and faces. Reserve full 360 orbits for longer durations on objects with simple materials and clean backgrounds.
Can I combine pan and dolly in one AI video prompt?
Short consumer clips on Kling, Pika, Luma, and Hailuo handle one camera path per generation. Combining pan and dolly in five seconds often yields diagonal mush. Generate two clips with separate moves and cut them in an editor, or extend duration on Runway Gen-4 before you test a compound path.
Do camera movement prompts work the same on Kling and Pika?
The verbs travel, but duration budgets differ. Kling and Runway Gen-4 often allow longer slow dollies and arcs. Pika short-form lanes reward compact lines with one move and explicit seconds. Rewrite travel angle down when you port a Runway orbit into a five-second Pika box.
How do I start if I only have a still photo?
Crop the still to the output aspect, upload it to an image-to-video host, and lead with preserve language plus one slow move or locked camera. If the still is soft or busy, run it through PromptMake /image first to draft a cleaner hero prompt for Midjourney or FLUX. Guests get about three /image runs per day; free registration raises that cap.
Are camera movement prompts the same as camera angle prompts?
No. Camera angle prompts for still images name height, lens, and shot size on a frozen frame. Camera movement prompts name how the lens travels across time in a video host. You can use both ideas in a pipeline: lock angle on a still, then add pan, dolly, or orbit when you animate.
Ready to generate your own prompts?
Free. No sign-up required. Works with all major AI models.