MiniMax H3 Prompt Guide: How to Write Better AI Video Prompts & Practical Examples
"Summary: This MiniMax H3 prompt guide covers prompt structure, text-to-video and image-to-video examples, common mistakes, and practical tips for improving AI video prompts."
Writing prompts for AI video generation can be tricky. A short prompt may create a usable scene, but it may not give you enough control over details such as character movement, camera work, or facial expressions.
To see how prompt wording affects the result, I tested several MiniMax H3 prompts on PicLumen, including text-to-video and image-to-video generation. This guide shares what I learned from these tests, along with practical prompt examples and ways to refine your prompts.
What Is MiniMax H3?
MiniMax H3, also known as Hailuo 3.0, is a general-purpose omni-modal video generation model developed by MiniMax. It can work with text, images, video, and audio as part of a unified generation workflow. It can also generate video with native stereo audio.
For video generation, H3 supports text-to-video as well as several reference-based workflows. These include generating from a first or last frame, connecting two keyframes, and using multiple image, video, and audio references. MiniMax lists output durations of 4–15 seconds and resolutions up to 2K.
This multimodal approach also changes how you can write prompts. A reference image can provide visual information about a character or scene, while the text prompt can focus on what happens next. MiniMax's own video prompting guide recommends treating an image as the starting state and then describing the visual path that follows it.
How to Write Effective MiniMax H3 Video Prompts
A useful starting structure is:
Subject + Action + Environment + Camera + Visual Style + Lighting + Sound
You do not need to include every element in every prompt. Focus on the details that matter to the shot.
Core Components of a Good AI Video Prompt for MiniMax H3
1. Define the Main Subject
Start by telling H3 what the main subject is.
For a person, useful details can include appearance, clothing, hairstyle, or other features that should remain consistent.
For example:
A young woman with shoulder-length dark hair, wearing a beige trench coat, walks through a rainy city street.
This is more specific than:
A woman walks outside.
For image-to-video generation, however, you may not need to repeat every visible detail from the reference image. The image can already establish much of the subject's appearance.
2. Describe the Action
Video prompts need to describe movement, not only appearance.
Compare:
A woman in a café.
with:
A woman sits by the window, lifts a coffee cup, takes a sip, and looks outside.
The second prompt describes the action more clearly.
For an important or complex movement, it can also help to describe the starting state and how the movement develops.
A useful structure is:
Starting State → Action → Change → Final State
For example:
A chef places a pizza on a wooden table, slices it slowly, and lifts one piece toward the camera.
This gives the model a clearer sequence to work with.
3. Control the Camera
A word such as “cinematic” describes the overall look, but it does not tell the model how the camera should move.
Use direct instructions such as:
The camera slowly pushes in toward the subject.
The camera tracks beside the woman as she walks.
The camera slowly pulls back from a close-up to a wide shot.
MiniMax's official prompt guide also recommends describing camera movement as part of the shot, including movements such as push in, pull out, tracking, pan, tilt, and arc.
4. Describe the Environment and Lighting
The environment establishes where the action takes place.
Depending on your scene, you can describe:
Location
Time of day
Weather
Important background elements
Lighting
Color palette
For example:
A quiet train station at sunrise, with warm sunlight entering through the tall windows and long shadows across the floor.
You do not need to describe every object in the scene. Focus on elements that affect the shot.
In one of my image-to-video tests, the reference image had a plain background, but the generated result introduced an outdoor environment even though I did not ask for a background change.
When the background is important, consider stating it directly:
Keep the original background unchanged while animating the character.
This is a prompt suggestion rather than a guarantee, so the output should still be reviewed after generation.
5. Define the Visual Style
Add a style description when the look of the video matters.
For example:
live-action cinematic
documentary style
realistic commercial photography
3D animation
watercolor animation
vintage film
MiniMax's prompting guide also uses styles such as cinematic, live-action, 2D animation, 3D CG, claymation, watercolor, and vintage film as examples.
When you use an image reference, however, the image may already provide much of the visual style. In that case, you can use the text prompt mainly to describe movement and scene development.
6. Add Sound When It Matters
MiniMax H3 can generate dialogue, music, ambience, and sound effects together with video.
You can describe sound naturally in your prompt:
Soft rain and distant traffic can be heard in the background.
Or:
Footsteps echo through the empty hallway, followed by the sound of a metal door closing.
For a more detailed scene, think about:
Dialogue — what the character says.
Sound effects — physical sounds such as footsteps, rain, or impacts.
Ambient sound — environmental sounds such as traffic, wind, or room ambience.
MiniMax's own guide separates these elements when describing the audiovisual timeline.
Common Prompt-Writing Mistakes for MiniMax H3 Users
1. Using Disconnected Keywords
A prompt such as:
Woman, city, rain, cinematic, realistic, neon.
gives the model several ideas, but does not clearly explain how they should work together.
A more useful prompt is:
A young woman walks slowly through a rainy city street at night. Neon signs reflect on the wet pavement while the camera tracks beside her.
The second version connects the subject, action, environment, and camera movement.
2. Asking for Too Many Actions at Once
A short video has limited time. If a prompt contains too many major actions, some of them may not be completed as intended.
For example:
A man enters a café, talks to three people, orders coffee, checks his phone, leaves, gets into a car, and drives away.
A more focused version is:
A man enters a quiet café and walks toward the counter. He places his phone on the counter and orders a coffee while the camera slowly follows him.
For a short clip, focusing on fewer connected actions can make the sequence easier to control.
3. Using Vague Camera Instructions
Words such as:
cinematic camera
or:
professional camera movement
do not clearly describe the shot.
Instead, specify the movement:
The camera slowly tracks beside the runner.
The camera pushes in from a full-body view to an upper-body shot.
This makes the intended camera movement more explicit.
4. Describing an Important Movement Too Generally
This was one of the clearest findings from my own testing.
In an image-to-video test, I used:
She slowly turns toward the camera and smiles.
The generated character did not exactly follow the movement I had in mind. Instead of clearly starting from a turned-away position and rotating toward the camera, she moved from a side-facing position toward the front.
I then refined the prompt:
At the beginning of the video, she is turned slightly away from the camera. She slowly rotates her body until she faces the camera, then turns her head to look directly into the lens. She gives a subtle, natural smile and maintains eye contact with the camera.
The revised version produced clearer body and head movement, and the smile appeared more natural. However, the character maintained eye contact throughout the clip instead of looking away first.
This suggests that when a specific movement matters, describing the starting state and movement sequence can provide clearer direction than using one broad action phrase.
5. Leaving Important Objects or Interactions Unspecified
In my test, the model introduced details that were not explicitly included in the prompt.
In one of my text-to-video tests, the simple prompt:
A young woman walks through a city street at night.
produced a rainy scene and added umbrellas, even though rain and umbrellas were not mentioned.
One background pedestrian also developed an unnatural umbrella interaction. The umbrella appeared later in the scene, but it was not clearly connected to the person's hand.
An unexpected umbrella interaction observed in a MiniMax H3 test
When an object or interaction is important, describe it directly.
For example:
The pedestrian holds the umbrella firmly in one hand while walking across the street.
This gives the model more information about the intended interaction.
Practical AI Video Prompt Examples for MiniMax H3
The following examples can be used as starting points for different AI video projects.
Text-to-Video Prompt Samples for MiniMax H3
Example 1: Basic vs. Detailed Prompt
One of my tests compared a simple prompt with a more detailed version.
Basic Prompt
A young woman walks through a city street at night.
Detailed Prompt
A young woman with shoulder-length dark hair, wearing a beige trench coat, walks slowly through a rainy city street at night. Neon signs reflect on the wet pavement. The camera tracks beside her at a slow speed. Soft rain and distant traffic can be heard in the background.
 |  |
Basic vs. detailed MiniMax H3 prompts tested on PicLumen
Both versions generated a coherent walking scene. The basic prompt was enough to establish the overall scene, while the detailed version gave clearer instructions for the subject, environment, camera movement, and sound.
The basic version also introduced an unexpected umbrella on a background pedestrian, and the umbrella interaction looked unnatural. The detailed version did not show an obvious issue of this kind.
The key is to add relevant details, not simply more words.
Example 2: Cinematic City Scene
Live-action cinematic night scene. A young woman in a dark trench coat walks slowly through a rainy downtown street, holding a transparent umbrella. Neon signs reflect on the wet pavement. The camera tracks beside her at a slow speed. Soft rain and distant traffic can be heard in the background.
Prompt focus: cinematic scenes, social videos, and atmospheric clips.
Example 3: Product Advertisement
Cinematic product commercial featuring a minimalist black smartwatch on a reflective black surface. A narrow beam of light moves slowly across the watch, revealing its metallic edges and glowing screen. The camera makes a slow arc around the product before pushing in toward the display. Clean studio lighting, subtle reflections, premium commercial style.
Prompt focus: product videos, e-commerce content, and promotional clips.
Example 4: Food Video
Close-up cinematic shot of a bowl of ramen on a wooden table in a warm Japanese restaurant. Steam rises from the soup as chopsticks slowly lift the noodles. The camera gently pushes in toward the noodles. Warm window light, shallow depth of field, realistic food textures. Soft restaurant ambience and subtle food sounds.
Prompt focus: food content, social media clips, and short advertisements.
You can also explore our Seedance 2.5 Prompt Guide for more AI video prompt examples and structured templates.
Image-to-Video Prompt Samples for MiniMax H3
For image-to-video generation, the reference image already provides much of the visual information. The text prompt can therefore focus on motion, changes, and camera behavior.
Example 1: Character Movement
Reference image used for the MiniMax H3 image-to-video test
Prompt
Keep the character's appearance and clothing consistent with the reference image. She slowly turns toward the camera and smiles. Her hair moves gently in the wind. The camera slowly pushes in.
Test Result
The character's appearance and clothing remained consistent with the reference image, and the camera produced a clear push-in from a full-body view toward an upper-body shot. However, the character did not complete the intended turning motion exactly as described. Her smile also became more exaggerated than expected.
The character remained consistent, but the turning motion and facial expression differed from the intended result.
The test showed that the reference image preserved the character's main appearance, while the prompt had more influence on her movement and camera behavior.
Example 2: Portrait Animation
Keep the character's appearance, hairstyle, and clothing consistent with the reference image. She slowly turns her head toward the camera and gives a subtle smile. Her hair moves gently in the wind while the camera remains mostly static.
Prompt focus: character consistency + facial movement + subtle motion.
Example 3: Landscape Animation
Preserve the composition and visual style of the reference image. Clouds move slowly across the sky while trees and grass sway gently in the wind. The camera gradually pushes forward toward the mountains. Keep the main landscape structure unchanged.
Prompt focus: environmental movement + camera movement + scene consistency.
Example 4: Product Animation
Keep the product shape, material, logo placement, and studio setup consistent with the reference image. The camera slowly moves from left to right around the product while soft highlights travel across its surface. The product remains centered throughout the shot.
Prompt focus: product consistency + camera movement + lighting.
My Key Takeaways from Testing MiniMax H3 Prompts
After comparing several prompt styles and reviewing the generated results, I found a few practical points worth remembering.
1. More detail can give you more control
In my first comparison, both the basic and detailed prompts generated a usable scene.
However, the detailed prompt gave clearer instructions about the subject, environment, camera movement, and sound. This made the intended shot easier to define.
At the same time, this does not mean that longer prompts are always better. The useful details are the ones that directly affect the scene.
2. Describe important movements as a sequence
A general instruction such as:
She slowly turns toward the camera.
may leave room for different interpretations.
In my image-to-video test, the character moved from a side-facing position toward the front, but she did not clearly start from a turned-away position and rotate toward the camera as I had intended.
I then made the movement more explicit by describing the starting position, body rotation, and head movement separately.
Refined prompt:
At the beginning of the video, she is turned slightly away from the camera. She slowly rotates her body until she faces the camera, then turns her head to look directly into the lens. She gives a subtle, natural smile and maintains eye contact with the camera.
The revised prompt produced clearer body and head movement, and the facial expression also appeared more natural. However, the character maintained eye contact throughout the clip instead of changing her gaze during the turn.
For important actions, I found it useful to think about the sequence:
Starting State → Movement → Final State
This can make the intended action more specific and give you a clearer starting point for further refinement.
3. Refine the part that needs fixing
My tests also showed that the first generation does not always match the intended result exactly.
Instead of rewriting the entire prompt immediately, it can be more useful to identify the specific problem first.
For example:
Problem: the character does not clearly turn toward the camera.
Then revise the relevant instruction:
Describe the starting position and body rotation more explicitly.
This also makes each revision easier to evaluate.
4. Similar prompts can still produce different details
Another thing I noticed is that not every difference between two generations can be explained by a prompt change.
For example, the hair movement in my revised image-to-video generation was more noticeable even though the wording about the hair remained unchanged.
The background treatment also differed between generations using the same reference image.
So I would treat prompt writing as a process of testing and refinement rather than a fixed formula.
Troubleshooting: Common Prompt-Related Issues & Fixes for MiniMax H3
Problem | What I Observed | What to Try |
|---|---|---|
The action is not executed exactly | The character performs the general movement but not the intended sequence | Describe the starting position, movement, and final state |
A facial expression feels exaggerated | A simple smile can become broader than intended | Use specific wording such as “a subtle, natural smile” |
The camera movement is unclear | The desired movement may not be obvious without an explicit instruction | Describe one camera movement directly |
The background changes unexpectedly | A new environment may appear even when no background change was requested | State clearly whether the original background should remain unchanged |
An object interaction looks unnatural | An object may appear without a clear connection to the subject's action | Describe how the subject holds, touches, or moves the object |
Some details vary between generations | Hair or background treatment can differ even with similar prompt instructions | Review the output before assuming the prompt is the only variable. |
Practical Workflow: Use MiniMax H3 on PicLumen
PicLumen integrates MiniMax H3 into its AI Video Generator, with support for text prompts, frame-based generation, and multimodal references.
Brief Intro to MiniMax H3 Key Features
With MiniMax H3 on PicLumen, you can:
Generate videos from text
Animate a first or last frame
Connect two keyframes
Use image, video, and audio references
Generate video with native stereo audio
Overview of MiniMax H3 on PicLumen
Step 1: Select MiniMax H3
Open the PicLumen AI Video Generator and choose MiniMax H3 (Hailuo 3.0) from the model list.
Step 2: Add Your Prompt or References
Enter your prompt and upload a reference image, video, or audio file when needed. Assign a clear role to each reference.
Step 3: Generate and Review
Set the available video options, click Generate, and review the result. If the output does not match your intention, refine the specific part of the prompt that needs improvement.
Conclusion
Writing effective MiniMax H3 prompts starts with clear instructions about the subject, action, camera movement, and other details that matter to the shot. When a result does not match your intention, refining the specific instruction can be more useful than simply making the entire prompt longer.
You can test these MiniMax H3 prompt ideas on PicLumen and refine your results based on your own creative needs.
FAQ
What components should be included in prompts for MiniMax H3 video generation?
A useful MiniMax H3 prompt can include the subject, action, environment, camera movement, visual style, lighting, and sound.
You do not need every component for every video. Focus on the details that have the most impact on the shot.
How to structure an effective MiniMax H3 prompt?
Start with the main subject and action, then add the details that matter to the shot. Review the result and refine the part that needs improvement.
How to create effective prompts for MiniMax H3 video generation?
Start with one clear idea and describe the main subject and action first.
Then add the environment, camera movement, style, lighting, and sound when they are relevant.
A simple starting formula is:
Subject + Action + Environment + Camera + Style + Sound
Review the result after generation and refine the parts that do not match your intention.
What prompt techniques improve video output quality from MiniMax H3?
Based on my testing, clear instructions were more useful than simply adding more words.
Describing important actions as a sequence also helped me define the intended movement more clearly. For image-to-video generation, I found it useful to let the reference image carry the visual details and use the text prompt to focus on motion and changes.
These are practical observations from my tests, not guarantees for every generation.
How to generate prompts from an existing video for MiniMax H3?
Start by breaking the video into its main elements:
Subject
Action
Camera movement
Scene changes
Important sounds
Then rewrite these elements in chronological order.
For example:
A man walks through a narrow hallway toward a closed door. The camera tracks backward in front of him. He stops, reaches for the handle, and opens the door as warm light enters the hallway.
MiniMax H3 also supports video references as part of its multimodal workflow, so an existing clip can be used as an additional source of motion or scene information.
Where can I test MiniMax H3 for AI video generation?
You can test MiniMax H3 directly on PicLumen. The platform currently offers text, frame-based, and multimodal reference workflows for H3.
