Camera control · Prompt builder

MiniMax H3 Camera Control

Camera control in H3 happens entirely in the prompt text, and the model only recognises a fixed set of words for it. This page lists all 20 motion types MiniMax publishes, shows how amplitude and speed change a shot, and writes the camera control sentence for you.

20 official motion typesAmplitude + speed modifiersRuns in the generator

Camera prompt builder

Describe the move, get the shot

Pick a motion type and the builder writes the camera control clause the way the guide specifies — verb, amplitude, speed, and the target the move lands on. The output is a complete T2VA prompt you can run as-is.

Your prompt

English, in the three-field format H3 expects.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, a medium-wide shot frames a watchmaker at a cluttered bench under a single warm desk lamp. The camera pushes in with small amplitude at slow speed toward the open watch case between his hands. He sets down the tweezers and turns the movement a quarter turn to catch the light.

overall_soundscape: A low room tone carries throughout, broken by the tick of the movement, the light clink of tweezers on wood, and a single chair creak

non_diegetic_music: Sparse piano notes at a slow tempo with a sustained string pad underneath
Run this prompt

Opens the generator with this prompt already filled in.

Prompt check

Nothing to flag — this prompt follows the guide.

Diagnosis

Why the camera ignores you

H3 has no camera widget and no joystick — camera control reaches the model only through the prompt text, which makes it a writing problem. Three habits account for most of the shots that come back wrong.

1

You invented the verb

MiniMax publishes a fixed vocabulary of 20 motion types. “Camera moves left” is not one of them, and neither is “drone shot” or “slow zoom out effect”. The model was trained on the published wording, so a synonym is a coin flip. Pick from the table below and paste the phrase as written.

2

The move has nothing to land on

Every camera example MiniMax prints ends on a target — what comes into frame, what the lens settles on, what the move reveals. Direction alone tells the model which way to go but not when to stop or what matters, so it drifts or barely moves.

3

The camera is buried in the scene description

The guide asks for camera motion written as a natural action inside the shot, not stacked as labels at the end of a sentence. A prompt that finishes with “, cinematic, pan right, 4k, dynamic camera” reads as style tags, and the move gets treated as decoration.

Reference

All 20 camera moves H3 recognises

Every camera control phrase H3 recognises, straight from section 4.3 of the prompt guide MiniMax ships with the weights. The phrase column is what goes in your prompt — copy it exactly.

The lens pivots or zooms, body still

Motion typeWrite it asWhat it does
Zoom InThe camera zooms inFocal length changes while the body stays put — the frame tightens without the camera travelling.
Zoom OutThe camera zooms outFocal length widens while the body stays put. Not the same as Pull Out, and the two look different.
Pan LeftThe camera pans leftThe camera stays in place and the lens pivots left.
Pan RightThe camera pans rightThe camera stays in place and the lens pivots right.
Tilt UpThe camera tilts upThe camera stays in place and the lens pivots upward.
Tilt DownThe camera tilts downThe camera stays in place and the lens pivots downward.

The whole camera moves

Motion typeWrite it asWhat it does
Static ShotThe camera holds a static shotCamera position and lens both stay still.
Push InThe camera pushes inThe camera body moves forward toward the subject.
Pull OutThe camera pulls outThe camera body moves backward away from the subject.
Truck LeftThe camera trucks leftThe whole camera slides left. Use this, not Pan Left, when you want the viewpoint itself to move.
Truck RightThe camera trucks rightThe whole camera slides right.
Pedestal UpThe camera pedestals upThe entire camera rises vertically.
Pedestal DownThe camera pedestals downThe entire camera drops vertically.

The camera relates to the subject

Motion typeWrite it asWhat it does
Arc ShotThe camera arcs around the subjectThe camera travels in an arc around the subject.
Tracking ShotThe camera tracks the subjectThe camera follows a moving subject.
POVThe camera holds the subject's point of viewThe frame becomes the subject's own point of view.

Roll and shake

Motion typeWrite it asWhat it does
Roll ClockwiseThe camera rolls clockwiseThe camera rotates clockwise around the lens axis.
Roll CounterclockwiseThe camera rolls counterclockwiseThe camera rotates counterclockwise around the lens axis.
Shake SlightlyThe camera shakes slightlyA light, continuous camera shake.
Shake StronglyThe camera shakes stronglyA heavy camera shake.

Pan versus Truck and Zoom versus Push are the two pairs people mix up most. Pan and Zoom keep the camera body in one place; Truck and Push move it, which changes the parallax and reads as a completely different shot.

Precision

Amplitude and speed

A complete camera control expression has three dimensions, not one: the motion type says how the camera moves, amplitude says how much the composition changes, and speed says how fast that change happens. Most prompts only ever supply the first.

Amplitude — how far the frame travels

“with small amplitude” keeps the composition close to where it started; “with large amplitude” swings it well away. This is the dial that separates a slight adjustment from a sweeping reveal, and it is independent of speed.

Speed — how fast it gets there

“at slow speed” and “at fast speed”. A large-amplitude move at slow speed and a small-amplitude move at fast speed are different shots entirely, and both are worth reaching for.

Leave out the middle

The guide is explicit that medium amplitude and normal speed are normally omitted. Writing them out adds words without adding instruction — and every unnecessary word in a camera clause competes with the parts that do carry meaning. Say nothing and you get the default.

Shape

Write it as an action, not a tag

Camera motion belongs inside the shot description as a natural English action. These are the three examples MiniMax prints in the guide, unchanged.

  • The camera pushes in with small amplitude at slow speed toward the folded letter in her hands.
  • The camera pans right with large amplitude at fast speed, revealing the open doorway.
  • The camera holds a static shot as the runner exits the frame.

Notice what all three camera control sentences have in common: motion type, then modifiers where they matter, then the thing the move arrives at.

1

Motion type first

The published phrase, verbatim: “The camera pushes in”, “The camera pans right”, “The camera holds a static shot”.

2

Modifiers only when they mean something

The first two examples carry amplitude and speed because the shot depends on them. The third omits both — a static shot has nothing to modify.

3

Then the landing

“toward the folded letter in her hands”, “revealing the open doorway”, “as the runner exits the frame”. This is what turns a direction into a shot.

Boundaries

What one shot cannot do

A single shot holds a single camera move. You cannot ask H3 to pan left for two seconds and then pan right for the rest — instructions that fight each other inside one shot tend to cancel out, and you get a camera that barely moves at all.

Use a cut instead

Two moves means two shots. The guide writes the cut with a timestamp, and the builder above adds it for you when you turn on the second shot:

[Shot 2] At 00:04.000, the camera cuts to a closer angle on the same subject. The camera arcs around the subject.

For anything past 15 seconds the same logic applies across segments — the long video planner splits a longer piece into shots and carries continuity between them.

And when the shot has to end on a specific frame rather than a specific move, that is a different prompt format — see the first and last frame prompt builder.

Longer single takes hold a move for longer without cutting — that is what the H3 Max tier is for.

And if you want the reasoning behind all of this rather than the reference table, why most camera movement prompts fail works through it with worked examples.

And when you want the camera placed by number rather than described in words — an exact orbit around a frozen scene — H3 Max Multi Angle takes camera keyframes instead of this vocabulary.

FAQ

Camera control questions

Can I control the camera in MiniMax H3 with keys or a joystick?

No. The hosted H3 endpoints take a text prompt and nothing else, so every camera instruction has to be written into the prompt. That is why the exact wording matters — the vocabulary on this page is the whole interface.

Why does “camera moves to the left” not work reliably?

It is not in the published vocabulary, and it is also ambiguous: it could mean Pan Left (the lens pivots) or Truck Left (the whole camera slides). Those are different shots. Naming the motion type removes both problems.

Can the camera change direction in the middle of a clip?

Not inside one shot. Cut to a second shot at a timestamp and give that shot its own motion type — the builder above writes the cut for you. Stacking two directions into a single shot usually produces a camera that hardly moves.

Do amplitude and speed work on every motion type?

No. Static Shot and POV describe a state rather than a movement, and Shake Slightly and Shake Strongly already carry their own degree. The guide defines modifiers for the other sixteen.

Does camera vocabulary work the same in image-to-video?

The vocabulary is identical, but the prompt around it changes: image-to-video and first/last-frame modes each open with their own reference instruction before the three core fields. The move itself is written exactly as it is here.

Where does this list come from?

Section 4.3 of the Video Prompt Writing Guide that MiniMax publishes alongside the H3 model weights. It is the model author’s own documentation, not a third-party prompt collection, and this page links to the source.

MiniMax Video Prompt Writing Guide · September 2026

Write the move, run the shot

The builder outputs a complete prompt in the format H3 expects, camera control clause included. Send it straight to the generator and see the move on real footage.

Open the builder

Looking for full prompts to start from? The prompt library has ready-made scenes you can adapt.