THE ULTIMATE MINIMAX H3 PROMPTING GUIDE
THE ULTIMATE MINIMAX H3 PROMPTING GUIDE
How to Write Better Video Prompts with Cinematic Shots, Camera Movements, Dialogue, Sound Effects and Music
1. Introduction – Understanding MiniMax H3
MiniMax H3 is an audiovisual generation model capable of creating video and synchronized audio from a single prompt.
Unlike traditional text-to-video models, H3 can generate several audio elements alongside the visuals:
Dialogue and spoken words.
Human vocalizations and reactions.
Environmental ambience.
Physical sound effects.
Background music and cinematic scores.
Sound events synchronized with specific actions.
This means that writing an effective H3 prompt requires a slightly different approach.
You are not simply describing what the camera should see. You are directing an entire audiovisual scene.
Think like a film director:
What do we see? What happens? How is it filmed? What do we hear? When do we hear it?
A good prompt should answer these questions clearly.
The following guide explains how to structure your prompts, control the action, describe camera movements, and create convincing soundscapes.
2. The Fundamental Structure of an H3 Prompt
For most simple Text-to-Video and Image-to-Video generations, a useful starting structure is:
Scene and environment.
Main subject.
Subject action and behavior.
Camera framing and movement.
Lighting and visual atmosphere.
Dialogue and vocalizations, if required.
Environmental sounds and sound effects.
Background music, if required.
You do not necessarily need to write eight separate sections for every generation. They are the eight elements to consider when building your description.
The most important principle is to describe events in a logical order.
Recommended basic template:
[SCENE]
Describe the location, environment and time of day.
[SUBJECT]
Describe the main character or object.
[ACTION]
Describe exactly what the subject does.
[CAMERA]
Describe the framing, angle and camera movement.
[VISUAL ATMOSPHERE]
Describe lighting, colors and visual style.
[DIALOGUE]
Describe spoken words and vocal delivery, if any.
[OVERALL SOUNDscape]
Describe environmental ambience and physical sounds.
[NON-DIEGETIC MUSIC]
Describe background music, if any.The section names above are a convenient writing aid. They are not all mandatory special commands.
For simple generations, natural language is perfectly suitable. For more elaborate scenes, the structured format helps keep the instructions organized.
3. Scene and Subject – Establishing What We See
Start by establishing the environment and the main subject.
Avoid vague descriptions such as:
A woman in a room.Instead, describe the scene with enough detail to establish its visual identity:
A cinematic live-action scene inside a luxurious modern office.
A confident businesswoman in her forties stands beside a large window.
Warm afternoon sunlight enters the room, illuminating the furniture
and creating soft shadows across the floor.This gives H3 several useful visual instructions:
The location.
The character's appearance and approximate age.
Her position.
The environment.
The lighting conditions.
The overall visual style.
Important: Avoid describing too many unrelated objects or characters. Every additional element competes for the model's attention.
For a single-character scene, establish one main subject and make the surrounding environment secondary.
4. Actions and Character Behavior
The action is one of the most important parts of a video prompt.
A video model needs to understand not only what the subject looks like, but also what happens over time.
Compare these examples:
Weak:
A woman standing in an office.Better:
A woman stands beside her desk, slowly turns her head toward the camera,
then takes two steps forward while maintaining eye contact.The second prompt describes a sequence of observable events.
A useful approach is to divide the action into three stages:
Beginning: What is the initial position?
Development: What movement or action takes place?
Ending: What is the final position or reaction?
Example:
A young woman is standing near a window.
She slowly turns around, looks directly at the camera
and takes a small step forward.
She finishes the movement with a subtle smile.This creates a clear progression.
4.1 Describing Natural Movement
Use precise verbs rather than generic expressions.
Examples:
Slowly turns her head.
Gently raises her hand.
Takes two deliberate steps.
Looks over her shoulder.
Briefly closes her eyes.
Shifts her weight from one leg to the other.
Adjusts her jacket.
Leans forward slightly.
Smiles naturally.
Walks toward the camera.
4.2 Avoiding Conflicting Actions
Do not ask a character to perform several complex actions simultaneously.
For example:
She walks forward, turns around, sits down,
looks behind her and raises both hands.This is difficult to interpret as a coherent movement.
A better version:
She walks slowly toward the chair.
She turns slightly, then sits down.
After sitting, she looks back over her shoulder.The sequence is much easier to understand.
5. Camera Language – Controlling the Shot
Camera direction is essential for cinematic results.
A common mistake is to describe only the character's movement while leaving the camera completely undefined.
The camera should be treated as an independent element.
There are two important distinctions:
Subject movement: What the character does.
Camera movement: How the camera moves relative to the scene.
For example:
The woman slowly walks toward the window.
The camera remains stationary.Here, the character moves while the camera stays fixed.
Compare this with:
The woman remains standing.
The camera slowly pushes in toward her face.Here, the character remains still while the camera moves closer.
5.1 Camera Framing
Common framing instructions:
Camera framingDescriptionExtreme close-upVery tight detail, such as eyes or lipsClose-upHead and faceMedium close-upHead and upper torsoMedium shotUpper body and waistMedium-wide shotMost of the body and surroundingsFull-body shotEntire subject visibleWide shotSubject and environmentExtreme wide shotLarge environment, subject relatively small
Examples:
A medium close-up frames her face and shoulders.A full-body shot shows the character from head to toe.5.2 Camera Angles
Camera angleDescriptionEye-levelCamera at the subject's eye heightLow angleCamera looking upwardHigh angleCamera looking downwardBird's-eye viewCamera directly aboveDutch angleCamera tilted sidewaysOver-the-shoulderCamera positioned behind a subject's shoulderPOVFirst-person perspective
Examples:
A low-angle cinematic shot looking slightly upward at the subject.A high-angle shot looking down at the woman sitting on a chair.5.3 Camera Movements
These are useful terms to include in your prompts:
Static camera
The camera remains completely stationary.
The camera remains locked in a fixed position.Slow push-in
The camera physically moves closer to the subject.
The camera slowly pushes in toward the woman's face.Pull-back
The camera moves backward, revealing more of the environment.
The camera slowly pulls back, revealing the entire room.Pan left / Pan right
The camera rotates horizontally.
The camera slowly pans to the right, following the subject.Tilt up / Tilt down
The camera rotates vertically.
The camera gently tilts upward from her shoes to her face.Tracking shot
The camera follows the moving subject.
The camera smoothly tracks alongside the woman as she walks.Orbit shot
The camera moves around the subject.
The camera slowly orbits around the character in a smooth semicircle.Dolly shot
The camera moves physically forward or backward.
A slow cinematic dolly-in toward the subject.Zoom-in / Zoom-out
The focal length changes, creating an optical zoom effect.
A gradual optical zoom-in toward her face.Important distinction: Push-in versus Zoom-in
A push-in physically moves the camera closer to the subject. A zoom changes the focal length without necessarily changing the camera position.
They are not interchangeable.
5.4 Combining Camera and Subject Movement
For better control, explicitly describe both movements.
Example:
The woman slowly turns toward the camera.
The camera remains stationary throughout the movement.Another example:
The woman remains seated and looks directly at the viewer.
The camera slowly pushes in toward her face,
creating an intimate cinematic close-up.If you want the camera to move while the subject remains still, say so explicitly.
6. Shot Planning and Multiple Shots
MiniMax H3 can interpret prompts containing multiple shots.
This is particularly useful when you want a small cinematic sequence instead of a single continuous camera movement.
A multi-shot prompt should describe each shot in chronological order.
Example:
[Shot 1]
A wide cinematic shot of a woman standing in a modern office.
The camera remains static as she looks toward the window.
[Shot 2] At 00:02.500
Cut to a medium shot.
She slowly turns toward the camera.
The camera begins a gentle push-in.
[Shot 3] At 00:04.000
Cut to a close-up of her face.
She smiles subtly while maintaining eye contact.The first shot does not require a timestamp. Subsequent shots use their cut times.
Important: A shot change is different from a camera movement.
A camera movement changes the framing within the same shot.
A cut introduces a new shot and a new composition.
Do not introduce unnecessary cuts when you want a continuous take.
For example:
A single uninterrupted cinematic shot.
The camera slowly tracks the subject without cuts.This is a useful instruction when continuity is important.
7. Sound Design – One of H3's Most Interesting Features
H3 generates audio together with video.
A good prompt should therefore describe the soundscape with the same care as the visual scene.
There are three main audio categories to understand:
Dialogue and vocal sounds.
Environmental ambience and physical sound effects.
Non-diegetic background music.
Understanding the difference between them makes your prompts considerably clearer.
8. Dialogue and Vocalizations
Dialogue is sound produced by a character within the scene.
For example:
A woman looks toward the camera and says in a calm, confident voice:
"Are you ready?"If you want a specific language, state it explicitly:
She speaks in French with a natural Parisian accent.You can also describe the delivery:
Whispering softly.
Speaking calmly.
Speaking with confidence.
Shouting angrily.
Laughing while speaking.
Speaking slowly and deliberately.
Hesitating before answering.
Example:
A man speaks in a deep, relaxed voice.
His delivery is slow and natural, with a slight pause between sentences.8.1 Dialogue Timing
When dialogue must occur at a particular moment, associate it with the corresponding action or shot.
Example:
[Shot 1]
A man enters the room and looks toward the camera.
[Shot 2] At 00:02.000
He stops, smiles and says in French:
"Bonjour, comment allez-vous?"For more structured prompts, the official H3 guide also documents the <d> dialogue notation and speaker identifiers such as (S1).
Example:
(S1) <d>French: Bonjour, comment allez-vous?</d>These conventions are particularly useful in complex multi-speaker scenes.
9. Overall Soundscape – Environmental Ambience
The overall soundscape describes the sounds that surround the scene.
These sounds establish the acoustic environment and help make a video feel realistic.
Examples:
Outdoor countryside:
overall_soundscape:
A gentle breeze moves through the trees.
Birds sing intermittently in the distance.
Soft rustling leaves and occasional footsteps
create a natural countryside atmosphere.Busy city street:
overall_soundscape:
Distant traffic noise, passing vehicles,
faint conversations and footsteps on the pavement.Indoor office:
overall_soundscape:
A quiet indoor room tone with a subtle ventilation hum.
Occasional keyboard typing and distant office activity.Rainy environment:
overall_soundscape:
Steady rainfall against the windows,
occasional distant thunder and water dripping from the roof.9.1 Physical Sound Effects
Physical sounds are associated with specific actions.
Examples:
ActionSound descriptionWalkingSoft footsteps on wooden flooringOpening a doorDoor handle clicks and hinges creakSitting downChair creaks slightlyPouring waterLiquid splashing into a glassRunningRapid footsteps and breathingWindFabric and leaves rustlingObject fallingA short impact followed by a rolling sound
Example:
As she walks across the wooden floor,
her heels produce distinct rhythmic clicking sounds.
Her clothing makes a subtle fabric rustling sound.These details should be included close to the action that produces them.
10. Non-Diegetic Music – Background Music
This is one of the most important distinctions in H3 audio prompting.
Non-diegetic music is music heard by the audience but not by the characters.
It is the equivalent of a film soundtrack.
For example:
Cinematic orchestral music.
Ambient electronic music.
Dark atmospheric drones.
Minimal piano.
Synthwave.
Industrial soundscapes.
Epic trailer music.
Subtle suspense music.
A useful music description should specify:
Musical genre.
Instruments and sound sources.
Tempo.
Rhythm.
Dynamic evolution.
Whether the music should remain subtle or prominent.
Example – Cinematic ambient:
non_diegetic_music:
A slow atmospheric electronic score with deep evolving synthesizers,
soft granular textures and distant reverberating pulses.
The music gradually builds without becoming aggressive.Example – Dark cinematic soundtrack:
non_diegetic_music:
A dark cinematic score featuring low orchestral strings,
deep sub-bass drones and sparse metallic percussion.
Slow tempo, restrained rhythm and gradually increasing tension.Example – Minimal piano:
non_diegetic_music:
A delicate solo piano composition with slow sustained notes,
subtle reverb and a restrained emotional progression.Example – Techno:
non_diegetic_music:
A dark hypnotic techno track with a steady four-on-the-floor kick,
deep rolling bassline, metallic percussion and evolving analog synthesizers.
Driving rhythm at approximately 126 BPM.10.1 Music That Evolves During the Video
You can describe changes in the soundtrack over time.
Example:
non_diegetic_music:
A minimal atmospheric electronic score begins with a soft low-frequency drone.
A subtle pulsing synthesizer gradually enters.
The rhythm becomes more pronounced toward the end,
finishing with a sustained atmospheric texture.This is more useful than simply writing "epic music" or "dramatic soundtrack".
Tip: Avoid describing too many musical instruments and simultaneous changes. A clear musical identity is generally more effective than an excessively complicated soundtrack.
11. Diegetic Music Versus Non-Diegetic Music
These two concepts should not be confused.
Diegetic music: Music that physically exists inside the scene.
Examples:
A radio playing in a room.
Music coming from a car stereo.
A musician playing a guitar.
A DJ performing in a club.
Non-diegetic music: Music added as a soundtrack for the audience.
Example:
non_diegetic_music:
A subtle cinematic ambient score with soft synthesizer pads.For music coming from a radio:
integrated_multimodal_description:
A small radio plays a quiet jazz melody in the background.This distinction helps H3 understand where the music belongs in the scene.
12. Complete Prompt Examples
The following examples combine visual direction, camera movement and audio instructions.
Example A – Cinematic Office Scene
[Shot 1]
Live-action cinematic scene inside a modern executive office.
A confident businesswoman in her forties stands beside a large window.
Medium-wide shot, warm afternoon sunlight and realistic skin textures.
She slowly turns toward the camera and takes one deliberate step forward.
The camera remains stationary.
overall_soundscape:
Quiet office ambience, subtle ventilation noise,
soft footsteps on the wooden floor and gentle clothing movement.
non_diegetic_music:
A restrained cinematic ambient score with soft piano notes
and warm atmospheric synthesizers.Example B – Countryside Atmosphere
[Shot 1]
A beautiful countryside scene during late afternoon.
A woman wearing an elegant floral summer dress stands in a green meadow.
Warm sunlight illuminates the landscape.
She gently turns her head toward the camera,
smiles and slowly walks forward.
The camera performs a smooth tracking shot,
following her movement.
overall_soundscape:
Gentle wind blowing through the grass,
birds singing in the distance and soft footsteps on the ground.
non_diegetic_music:
A delicate acoustic soundtrack featuring fingerpicked guitar,
soft piano and subtle orchestral strings.Example C – Dark Science Fiction
[Shot 1]
A futuristic industrial corridor illuminated by blue neon lights.
A lone character walks slowly through the environment.
Cinematic wide shot with atmospheric volumetric lighting.
The camera tracks backward in front of the character,
maintaining a constant distance.
overall_soundscape:
Deep mechanical ventilation, distant metallic impacts,
electrical humming and subtle machinery vibrations.
non_diegetic_music:
A dark futuristic electronic score with deep analog synthesizers,
sub-bass drones, slow rhythmic pulses and metallic textures.Example D – Techno Club
[Shot 1]
A cinematic underground techno club filled with atmospheric smoke,
red lighting and moving laser beams.
A DJ performs behind a professional mixing console.
The camera slowly pushes toward the DJ
while the lighting pulses with the rhythm.
overall_soundscape:
Crowd ambience, distant voices and subtle room reverberation.
non_diegetic_music:
Dark hypnotic techno at 128 BPM,
powerful four-on-the-floor kick,
deep rolling bassline, industrial percussion
and evolving analog synthesizers.13. Common Prompting Mistakes
Even a detailed prompt can produce inconsistent results if the instructions are contradictory.
Here are some mistakes to avoid.
Mistake 1 – Too many actions
A character cannot perform five unrelated movements simultaneously in a convincing way.
Solution: Describe a chronological sequence.
Mistake 2 – Undefined camera behavior
If you only describe the subject, H3 may choose its own camera movement.
Solution: Specify whether the camera is static, tracking, panning, tilting or pushing in.
Mistake 3 – Contradictory camera instructions
Example:
The camera remains static while continuously orbiting around the subject.These instructions conflict.
Solution: Choose one coherent camera behavior.
Mistake 4 – Vague audio descriptions
Example:
Beautiful cinematic music and realistic sounds.Solution: Specify the genre, instruments, rhythm and environmental sounds.
Mistake 5 – Mixing every audio category together
Dialogue, environmental sounds and background music serve different purposes.
Solution: Keep them conceptually separated.
Mistake 6 – Overloading the prompt
Long prompts are not automatically better prompts.
Too many visual, movement and audio instructions can create competing priorities.
Solution: Focus on the elements that matter most for the intended scene.
Mistake 7 – Forgetting sound continuity
If a scene contains continuous rain, wind or music, specify that it continues throughout the video.
Solution: Describe persistent sounds in the overall soundscape and localized sounds alongside their corresponding actions.
14. A Practical Prompting Workflow
For consistent results, use an iterative approach.
Step 1 – Establish the scene
Define the subject, environment and visual style.
Step 2 – Define the main action
Describe one principal action and its progression.
Step 3 – Choose the camera
Select the framing, angle and movement.
Step 4 – Build the soundscape
Add environmental ambience and physical sound effects.
Step 5 – Add dialogue or music
Only when necessary.
Step 6 – Generate and evaluate
Watch the video and listen to the audio separately.
Ask yourself:
Is the character performing the intended action?
Is the camera moving correctly?
Is the scene visually coherent?
Are the sound effects synchronized?
Is the music appropriate?
Is the soundtrack too loud or too dominant?
Step 7 – Change only what is necessary
If the visuals are correct but the audio is wrong, modify the audio instructions rather than rewriting the entire scene.
Likewise, if the soundscape is excellent but the camera movement is incorrect, focus on the camera instructions.
This makes experimentation more efficient.
15. Final Recommendations
The best MiniMax H3 prompts are not necessarily the longest ones.
They are the prompts that communicate a clear cinematic intention.
Think of H3 as a virtual film crew:
The scene description establishes the set.
The subject description defines the actors and objects.
The action describes the performance.
The camera instructions define cinematography.
The overall soundscape defines the acoustic environment.
The non-diegetic music defines the soundtrack.
Dialogue instructions define the spoken performance.
Once you understand these different layers, you can build much more convincing audiovisual sequences.
The key principle: Don't just describe what should happen. Describe how it should look, how it should move, and how it should sound.
Happy generating!