Camera Movement Prompts for AI Video (Zoom, Pan, Dolly)

Camera movement prompts work best when the move comes from a built-in chip and the wording matches the model family you picked. In VdoBloom, the Text to Video and Image to Video tabs ship camera-move chips such as push-in, orbit and dolly zoom, plus lens chips such as anamorphic and macro, and PixVerse models add 21 named movements of their own. The working rule is one move per clip, phrased the way that model reads direction.

Updated October 2026.

Where camera control actually lives

This is the part most guides skip. Camera direction is not purely a prompting problem. VdoBloom’s Text to Video and Image to Video tabs carry built-in camera-move chips — push-in, pull-back, orbit 360 degrees, crane up, whip pan, FPV one-take, dolly zoom, crash zoom, top-down, handheld, rack focus and static tripod — plus a set of lens or look chips: anamorphic, 85mm portrait, wide 24mm, 16mm film, macro, tilt lens, dreamy glow and 70mm epic. Selecting a chip is more reliable than typing the same phrase, because the chip is already phrased the way each model family expects.

Step by step: adding a camera move

Open the video creation dashboard and pick Text to Video if you are starting from a written prompt, or Image to Video if you have a still.
Choose the model you want to shoot with. The video picker filters by input type, so image-to-video models appear only after you have supplied an image.
Write the shot first: subject, action, setting, light. Leave the camera for its own sentence.
Open the camera-move chips and select a single move. Do not select two opposite moves.
Optionally add one lens or look chip, which is written into the prompt in the phrasing that selected model understands.
Check duration, resolution and aspect ratio, then read the credit cost shown before you generate.
Generate. If the move feels too slow or too fast for the clip length, shorten the duration rather than stacking a second move.

Camera moves and the phrasing each family expects

Move	What it does	How to phrase it
Push-in	Moves toward the subject, tightening attention	Seedance: a “Camera:” sentence. Kling, Runway and Wan: “The camera pushes in.”
Pull-back	Retreats to reveal context around the subject	Hailuo: [Pull out]. Veo: “slow dolly-out”.
Orbit 360 degrees	Circles the subject for a full reveal	“The camera orbits 360 degrees around the subject.”
Crane up	Rises vertically for scale	Hailuo: [Pedestal up]. Veo: “crane shot”.
Whip pan	Snaps sideways between two subjects	List it alone; combined with another move it reads as a hard cut.
FPV one-take	A continuous flying move through a space	“A single continuous FPV move.”
Dolly zoom	Background stretches while the subject stays fixed	PixVerse names it “Hitchcock”; elsewhere write “dolly zoom”.
Crash zoom	A fast snap onto a detail	Hailuo: [Zoom in]. PixVerse: a quick zoom variant.
Top-down	Overhead flat-lay framing	“The camera looks straight down.”
Handheld	Adds subtle instability and energy	Hailuo: [Shake]. Seedance: a plain “Camera: handheld.”
Rack focus	Shifts focus between two depths	“The camera racks focus from the foreground to the background.”
Static tripod	Locks the frame so action reads clearly	Hailuo: [Static shot].

The phrasing matters because families do not share one camera vocabulary. Seedance models get their own “Camera:” sentence in plain natural language. Veo models want cinematography terms such as “slow dolly-in” and “crane shot”, and lens chips become a “Lens: ...” line. Hailuo, MiniMax’s family, expects bracketed commands where one exists: [Push in], [Pedestal up], [Pan right], [Zoom in], [Shake], [Static shot], [Tracking shot] and [Pull out]. Kling, Runway and Wan models take a “The camera ...” sentence, and other models render lens looks as “Shot on ...”.

PixVerse models are the exception that proves the value of chips. They carry 21 named camera movements of their own: zoom in, zoom out, smooth zoom in, quick zoom in, quick zoom out, pan left, pan right, move left, move right, move up, move down, crane up, rotation, Hitchcock, super dolly out, whip pan, robo arm, follow left, follow right, fixed background and auto. When a model names a move, use the name.

Six camera-move prompt examples

Each of these is a complete prompt with the camera sentence in place and nothing contradictory around it.

Push-in, Runway style: “A brass key rests on a velvet tray under warm side light. The camera pushes in slowly until the key fills two thirds of the frame.”
Orbit, Seedance style: “A ceramic vase on a plinth in a bright studio. Camera: orbit 360 degrees around the vase at a constant height, soft even light throughout.”
Crane, Veo style: “A rooftop garden at dusk with warm string lights. Slow crane shot rising above the planter line, keep the city skyline in the lower third.”
Pull-back, Hailuo style: “A single lit window in a dark building. [Pull out] until the whole facade is visible, static otherwise.”
Rack focus, Wan style: “A coffee cup on a counter with a person blurred behind it. The camera racks focus from the cup to the person, no movement.”
Dolly zoom: “A cyclist at a standstill on a long road with buildings behind. A dolly zoom stretches the background while the cyclist stays the same size, one continuous move.”

Combining a move with a lens or look

One camera move plus one lens chip is a strong pairing. Two camera moves is not. A push-in with an anamorphic look, or a static tripod with an 85mm portrait look, reads cleanly because the model has a single spatial instruction and a single rendering instruction. A push-in combined with a pull-back gives the model two opposite instructions and produces drift. Add the lens chip after the camera chip, keep both short, and let the move do the storytelling. If you want one specific dramatic move rather than a generic one, the dedicated dolly zoom tool packages the effect for you.

Common mistakes

Two opposite moves in one prompt, such as push-in and pull-back. Pick one.
Naming a move the model has no vocabulary for. If the family uses bracketed commands, use them.
Asking for a long, complex move inside a 5-second clip. A 360 orbit needs room; a crash zoom does not.
Relying on prompt text when a chip exists. The chip is already phrased correctly for the selected family.
Stacking a camera move on top of heavy subject motion, which splits the model’s attention.
Forgetting that a look chip changes rendering, not movement — it will not rescue a bad move.

Most of these come down to giving the model one clear instruction, which is the same lesson behind these common AI video prompt mistakes.

How long should the shot be

Match the move to the duration. A static tripod, a rack focus or a crash zoom can land inside 5 seconds. A slow push-in or a gentle crane works at 5 to 10 seconds. A full 360-degree orbit or an FPV one-take wants the longest clip the model allows, so duration control becomes part of camera control.

Which model to pick for camera work

Pick PixVerse when you want a move by name from its 21-movement list. Pick Kling 3.0 when you want per-second duration control so a slow move has room to breathe. Pick Wan 3.0 when you want a long take of up to 30 seconds for an uncut travelling move. Budget the shot through pricing before a long take, since longer and higher-resolution clips cost more credits.

Frequently asked questions

What is the best camera movement prompt?

The best prompt pairs one camera move with a specific subject and light, using the phrasing the model expects, and leaves the camera in its own short sentence rather than mixing it into the action.

Can I use two camera moves in one prompt?

You should not. Two moves, especially opposite ones such as push-in and pull-back, give the model contradictory instructions and produce drift. Pick one move per clip.

Where are the camera-move chips in VdoBloom?

They are built into the Text to Video and Image to Video tabs. The same tabs also carry lens or look chips such as anamorphic, macro and 70mm epic.

How do I phrase a camera move for a specific model?

Use the family convention. Seedance takes a “Camera:” sentence, Veo takes cinematography terms, Hailuo takes bracketed commands, and Kling, Runway and Wan take a “The camera ...” sentence.

How long should a camera movement last?

Short moves such as a crash zoom suit 5 seconds, gentle push-ins and cranes work at 5 to 10 seconds, and full orbits or FPV one-takes want the longest clip available.

Open the video creation dashboard, write the shot, then add a single camera-move chip and one lens chip before you generate.