Model Prompting Guides
MiniMax H3 Prompting Guide
Master MiniMax H3 video prompting through prompt structure, reference-asset roles, camera, sound, complete examples, and generated results.
MiniMax H3 can understand text, images, video, and audio at the same time, and generate video with synchronized sound. A prompt needs to make the generation goal, the role of each reference asset, the shot timeline, the subject's action, the camera, and the sound explicit. The more specific your description and the more consistent the pieces are with one another, the more controllable the result.
Core Information in a Prompt
An H3 prompt needs to answer three questions:
| Core information | What to spell out | Example |
|---|---|---|
| Generation goal | Scene, subject, visual style, and the main event | A cool-toned perfume product short: a water droplet slides down the bottle, ending on the label |
| Reference-asset role | What each image, video, or audio clip is specifically responsible for | Image 1 provides the product's appearance; Video 1 provides the camera path only |
| Audio-visual timeline | Opening, action changes, camera, sound, and the ending frame | Droplet slides down → camera pushes in → focus shifts to the label → holds on a product close-up |
A complete prompt usually consists of six parts:
- Scene sequence: how the video begins, develops, and ends;
- Subject action: who does what, and how the action happens and changes;
- Reference role: what each image, video, and audio clip provides;
- Camera: shot size, angle, movement, and cuts;
- Sound: dialogue, ambience, synchronized effects, and music;
- Ending frame: where the shot finally rests, and what must stay unchanged.
Reference assets: [Image / Video / Audio N] handles [character, product, action, camera, sound, or style].
Scene sequence: [opening state] → [main action and changes] → [ending state].
Subject action: [who] [does what specifically], with action speed and reaction of [description].
Camera: [initial shot size and angle], [main camera movement], ending on [final composition].
Sound: [dialogue]; [ambience]; [sound effects]; [background music].
Keep unchanged: [character identity, product packaging, logo, wardrobe, scene structure, etc.].
Avoid: [errors that genuinely affect delivery].
Complete Prompt Examples
Product Hero Shot
Input: Image 1 · 8 s · 9:16
| Part | Content |
|---|---|
| Reference | Image 1 handles the accurate appearance of the product |
| Sequence | Droplet slides down → camera pushes in → label lands |
| Camera | Slow push-in, focus shifts to the label |
| Sound | Subtle ambient floor tone and a glass chime |
| Keep | Packaging proportions, logo, label, and color |
Image 1 is the product appearance reference; fully preserve the bottle shape, label layout, logo placement, cap, material, and color. Cold mist drifts slowly behind the bottle, and fine droplets slide down the glass surface. Open on a medium close-up, the camera pushes in slowly, focus moves from the droplet to the label, and it finally rests on a crisp, readable product close-up. Cool-toned side light, clean and restrained reflections. Sound includes subtle ambient floor tone and one clear glass chime. Avoid duplicated products, distorted logos, extra text, or watermarks.
Character and Action Reference
Input: Image 1 · Video 1 · 10 s · 16:9
| Part | Content |
|---|---|
| Reference | Image 1 handles the character; Video 1 handles action and camera only |
| Sequence | Enters frame → turns around → looks at camera |
| Camera | Low-angle follow, rising to a medium shot at the end |
| Sound | Footsteps, fabric rustle, indoor ambience, and low-frequency music |
| Exclude | Do not copy the character, wardrobe, or background from Video 1 |
Image 1 provides the character's identity, face, hairstyle, and wardrobe. Video 1 is referenced only for the walking rhythm and camera path; do not reference its character, wardrobe, or scene. The character walks into a concrete showroom, moves forward, turns once, and looks directly at the camera. A low-angle follow shot holds a steady speed, rising to a medium shot before the final pose. Keep the same character throughout. Sound includes clear footsteps, subtle fabric rustle, a wide indoor ambience, and a restrained low-frequency electronic score.
How to Write Reference Assets
After adding reference assets, describe in the prompt what each one is responsible for, item by item. Do not just write "reference Image 1" or "reference Video 1"; make clear what must be preserved and what should not be copied.
| Asset | Information it is suited to provide | Prompt focus |
|---|---|---|
| Character image | Face, hairstyle, body type, wardrobe, and demeanor | For a multi-person image, indicate which character to use |
| Product image | Bottle shape, material, color, label, logo, and text | List separately the packaging details that must be accurately preserved |
| Action video | Gait, dance, gestures, camera path, or editing rhythm | State which movements to reference only, and exclude the original character and scene |
| Audio | Timbre, dialogue, beat, music style, or sound effects | State whether to use it as-is or only reference its sound characteristics |
Image 1 provides the character's face, hairstyle, and wardrobe; Image 2 provides the scene and lighting; Video 1 references only the action rhythm and follow-shot path, without copying its character, wardrobe, background, or sound.
How to Write the Camera
A camera description includes shot size, angle, camera movement, and the purpose of the movement. Do not just pile up shot terms; explain how the camera moves around the subject and what needs to be visible after it moves.
| Shot | Function |
|---|---|
| Push In | The camera moves toward the subject, emphasizing an expression, product, or key detail |
| Pull Out | The camera retreats, revealing the full character, scene, or spatial relationship |
| Tracking | Keeps a speed and distance close to a moving subject |
| Arc Shot | Moves around the subject, showing its outline, material, and spatial depth |
| Pan | The camera position stays fixed while the shot turns horizontally to a new subject or piece of information |
| Pedestal | The entire camera rises or lowers, changing the height of observation |
| Handheld | Subtle natural shake, suited to documentary, street, and UGC textures |
| Static | A fixed frame that lets the subject's action become the visual focus |
| POV | Observes the action and environment from the position of a character's eyes |
Open on a medium close-up at a slightly low angle. The camera does a small arc along the left side of the product at a slow, steady speed, then pushes in toward the label so the logo and bottle material come into sharp focus, and finally rests on a front-facing product close-up.
How to Write the Sound
Describe dialogue, ambience, synchronized sound effects, and music separately, and state which action or shot each appears in. The sound content should stay synchronized with the on-screen events.
- Dialogue: state who is speaking, their emotion while speaking, and the complete line;
- Ambience: describe the persistent spatial sound, such as mall chatter, waves, rain, or a nighttime city floor tone;
- Synchronized effects: match sound to a specific action, such as a cap opening, soles landing, fabric rubbing, or liquid poured into a glass;
- Background music: state the music's style, instruments, and tempo, and its dynamic changes at key moments.
As the character opens the package, there is a clear paper-tearing sound and the subtle sound of nails touching the box. The room keeps a quiet, soft indoor ambience. She looks at the camera and says, in a natural, excited but not overly exaggerated tone: "The texture is even better than I imagined." The background adds a light, clean electronic beat that intensifies slightly when the product close-up appears.
Making Prompts More Stable
- Match the number of actions to the duration: a short video concentrates on one main action; when several events are needed, split them in chronological order;
- Use observable verbs: change "very premium" into "cool side light moves along the glass edge, with clean, restrained reflections";
- Avoid camera conflicts: do not require static, a fast push-in, and a large arc at the same time;
- State the items to keep: character identity, wardrobe, packaging, logo, color, and key props need to be stated separately;
- State the ending frame: write where the shot finally rests and what must be clearly visible in the final frame;
- Keep only necessary constraints: focus constraints on problems that genuinely affect delivery, such as typos, distortion, duplicated products, and watermarks.
If the result deviates from the prompt, check in turn whether the subject, action, scene, and camera conflict; whether each reference asset carries only a clear role; whether the number of actions suits the video's duration; and whether the ending frame and items to keep are already spelled out.
Showcase: References and Generated Results
Text Generation: Hand-Drawn Kitchen Creature
15 seconds, 16:9 landscape. Blend live-action footage of a small kitchen at dusk with glowing hand-drawn animation. The kitchen has an old wooden table, a half-washed cup, a slightly fogged glass bottle, and a hanging dish towel. The camera looks like a hurried phone shot held in one hand: slight shake, hesitant close focus, backlit exposure breathing, and coarse grain in the shadows. Use room floor tone, fabric rustle, a soft cup clink, dripping water, the shooter's footsteps and breathing, and the faint electronic hum and tiny chirps of the hand-drawn little creature. Avoid huge eyes, fangs, threatening motion, sudden black screens, or jump-scare shots.
Generated resultHand-Drawn Kitchen Creature
Download MP4First-Frame Generation: Space Opera Trailer
Image 1opening frame and overall visual style
An epic theatrical space opera trailer. Keep a fast pace and grand scale, using crisp hard cuts, a shaking command bridge, white-hot flashes, extremely short black frames, and violent jump-drive impacts. The title uses a wide-tracked cinematic typeface with restrained material texture, subtle lighting, and a thin edge light. The text emerges from deep-space shadow, sweeps past a beam of starlight, and gradually widens its tracking, leaving a brief afterimage before flashing in a black frame.
Generated resultSpace Opera Trailer
Download MP4Multi-Reference Generation: Street Dance Motion Transfer
Image 1character appearance reference
Image 2character appearance reference
Video 1street dance motion reference
Download MP4Use Video 1 as the motion reference for the street dance performance; use Image 1 and Image 2 as the character appearance references.
Generated resultStreet Dance Motion Transfer
Download MP4Seedance 2.0 Complete Prompting Guide
A practical guide to multimodal references, video editing, extensions, shot planning, subject consistency, camera direction, sound, and production-ready Seedance 2.0 prompts.
Veo3 / Gemini Omni Prompt Guide
How to write stable 8-second video prompts for Veo3 / Gemini Omni and avoid common moderation or generation failures.