MiniMax H3 Prompt Guide: Practical Video Examples
Learn how to write MiniMax H3 prompts for text-to-video, first-and-last-frame, and reference-to-video workflows through practical examples.
Run MiniMax H3
Choose the workflow by the control you need. Prompt-only generation leaves room for invention. First-and-last-frame fixes composition. Reference-to-video carries identity, movement, style, or sound from supplied media.
| Workflow | Inputs | Best use |
|---|---|---|
| Text to Video | Prompt only | Original concepts and animation |
| First & Last Frame | One or two exact frames | Posters, key art, and transitions |
| Reference to Video | Prompt plus mixed media | Identity, motion, style, voice, and editing |
Prompting Examples
These examples cover commercial production, native multimodal reference, and precise editing. Each one shows the level of direction and the matching endpoint.
Production Work Across Commercial Use Cases
Brand films, motion design, short drama, ecommerce, interactive media, and stylized character work.
Brand Films & Cinematic Content
Campaign films, trailers, fashion pieces, and cinematic brand storytelling.
Vintage Binocular Brand Film
Reference to VideoPrompt
Assign the architecture, camera language, type, and logo to separate references. Use fast binocular scans, optical blur, exposure flicker, and hard cuts that settle cleanly into focus.
Epic Space Opera Teaser
Text to VideoPrompt
Create a fast theatrical space-opera teaser with an immense scale, a distressed command deck, sharp edits, wide-tracked titles, and a violent jump-to-warp impact.
Sci-Fi Mystery Teaser
Reference to VideoPrompt
Use one reference for the protagonist and another for the station. Preserve identity as warning lights, distant machinery, and a low signal build the mystery.
Desert Fashion Campaign
Reference to VideoPrompt
Assign talent, styling, vehicle, and landscape references separately. Cut between silhouettes, heat shimmer, moving fabric, and wide desert frames in one editorial palette.
Cyber-Grunge Fashion Film
Reference to VideoPrompt
Match the reference typography, halftone texture, photocopied edges, wardrobe, and acid accents. Keep the edit quick and use hard cuts only.
Visual Concepts & Motion Design
Title sequences, illustrated worlds, music-video language, compositing, and animated posters.
Retro Anime Crime Title Sequence
Reference to VideoPrompt
Treat every reference as a specific title card, character, prop, or color cue. Use split screens, print texture, readable type, and decisive graphic cuts.
Hand-Drawn Kitchen Creature
Text to VideoPrompt
A loose ink-and-watercolor creature sneaks through a kitchen and hides after a sudden sound. Preserve line wobble, paper texture, and handmade timing.
Neon Laundromat Encounter
Text to VideoPrompt
Film a late-night apparition as unstable phone footage with shake, glass reflections, exposure changes, and delayed autofocus under fluorescent light.
Cyber-Grunge Rap Music Video
Reference to VideoPrompt
Match the print texture and typographic attitude. Combine coarse grain, gate weave, scan misregistration, rough edges, and hard edits with a readable performance.
Green-Screen to Fairytale Composite
Reference to VideoPrompt
Remove the green background and rebuild the supplied environment. Match subject lighting, contact shadows, camera motion, and background response.
Animated Gallery Poster
First & Last FramePrompt
Animate the artwork without changing its border, frame, palette, figure style, or layout. Reveal text with a light type-on sound.
AI-Native Storytelling
Short drama, period mystery, and character-led vertical storytelling.
Snowy Bamboo Wuxia Mystery
Reference to VideoPrompt
Stage a cold nocturnal investigation among bamboo, mist, and snow. Lock faces and period wardrobe, favor measured close coverage, and exclude modern elements.
Vertical Family Confrontation
Reference to VideoPrompt
Create a grounded 9:16 family argument in a lived-in interior. Build emotion through tight pacing, controlled eyelines, and natural medium close-ups.
Vampire Romance Short Drama
Reference to VideoPrompt
Use separate references for both leads and the castle, preserve identity, and emphasize eye contact, danger, resistance, and dark romance without gore.
Product & E-Commerce Marketing
Product demonstrations, premium campaigns, and feature-led advertising.
Futuristic Eyewear Campaign
Reference to VideoPrompt
Use the portraits for identity and styling, then the product image for the exact eyewear silhouette. Match a severe white-studio luxury campaign.
Ergonomic Chair Product Film
Reference to VideoPrompt
Reveal the chair through a full rotation, mesh macro shots, airflow, lumbar animation, and adjustment demos while preserving product geometry.
Digital Experiences & Game Concepts
Interface motion, product pages, gameplay, and interactive experience concepts.
Interactive Game Equipment UI
Reference to VideoPrompt
Use character and UI references for a timed equipment-selection sequence with real labels, component swaps, confirmation, loading, and world entry.
Nike-Style Product Landing Page
Reference to VideoPrompt
Animate a performance-product page with oversized italic type, carbon textures, quick scrolling, scale interactions, and a fixed supplied product.
Automotive Website UI Animation
First & Last FramePrompt
Move the headline down, bring the copy panel up, and change the car lights from dark to red while preserving final alignment.
Rotating Product Page Reveal
First & Last FramePrompt
Reveal the interface from top to bottom, move upper and lower type in opposing directions, then rotate the product after the layout settles.
Animation & Stylized Visuals
Stylized action, character promos, first-person gameplay, and visual-novel transitions.
Claymation Lava Canyon Leap
First & Last FramePrompt
A clay fox sprints and leaps across a lava canyon in heroic slow motion while the camera races beneath its body to reveal the depth.
Fantasy Wuxia Character Film
Reference to VideoPrompt
Lock hair, crown, ribbons, layered fabric, sash, fastener, and tassels. Follow the storyboard in order and reserve facial views for close-ups.
Otome Male Lead Character Promo
Reference to VideoPrompt
Use the portrait as a strict identity lock for face, hairstyle, proportions, costume materials, and polished otome-CG rendering.
First-Person Tactical Gameplay
First & Last FramePrompt
Simulate eye-level gameplay with player sway, directional checks, firing, recoil, and continued movement in a cool, smoky military environment.
Interactive Otome Game Transition
First & Last FramePrompt
Treat both uploads as exact endpoints and bridge them with readable visual-novel choices, dialogue, refined UI motion, and restrained tension.
Character-Driven Romantic Scene
Reference to VideoPrompt
Use the male lead and interior references for one controlled close shot. Direct fabric, breathing, room tone, crystal, ice, and partial framing.
Native Multimodal Reference
One request can carry identity from a portrait, movement from footage, art direction from a mood board, and voice from a recording.
Multi-Asset Reference
Combine text, images, video, and audio in one generation.
Multi-Asset Cinematic Remix
Reference to VideoPrompt
Assign six images as named assets, then use the reference cut for shot rhythm, transition language, and music. State which file controls every creative role.
Live-Action Voxel Transformation
Reference to VideoPrompt
Keep buildings and people photoreal while converting only trees and cars into voxel objects. Preserve real motion, shadows, and transmitted light.
Macro Coffee-to-Fluid Transition
Reference to VideoPrompt
Push into coffee foam until its texture matches desert ridges, then continue into the landscape as one seamless shot without tears or visible compositing.
Character, Motion & Camera Reference
Transfer identity, performance, movement, framing, and timing.
Character Swap with Performance Reference
Reference to VideoPrompt
Replace the performers while matching expressions, timing, blocking, handoffs, reactions, and the playful foam exchange from the source clip.
Street Dance Motion Transfer
Reference to VideoPrompt
Use the video only for dance motion and timing, and the images for performer identity. Preserve foot placement, rhythm, camera path, and ground contact.
Capybara Motion Recreation
Reference to VideoPrompt
Replace three performers with photoreal capybaras while matching every drop, roll, jump, and final pyramid from a locked wide camera.
Voice Cloning & Transfer
Use a supplied recording to guide a new vocal performance.
Voice Clone Dialogue Transfer
Reference to VideoPrompt
Give the character new dialogue while matching the reference voice, cadence, and delivery. Keep mouth motion, expression, and timing aligned.
Precise Multimodal Editing
Use an existing clip as the stable base, name every change, and state which surrounding details must remain untouched.
Character & Object Editing
Replace, remove, or add subjects while preserving the source clip.
Cat-to-Dog Replacement
Reference to VideoPrompt
Replace only the cat with a dog. Preserve camera motion, timing, environment, lighting, occlusion, and all unrelated details.
Add a Character in Sync with the Others
Reference to VideoPrompt
Add one performer on the left in the same uniform and synchronize them with the group without changing existing people or camera movement.
Subject & Wardrobe Replacement
Reference to VideoPrompt
Replace the rear child with the referenced dog and swap one jacket for the supplied denim piece. Protect everything else.
Scene & VFX Editing
Change environment, time of day, lighting, and composited elements.
Green-Screen Environment Replacement
Reference to VideoPrompt
Remove the green screen, rebuild the target environment, relight the subject, and make background elements respond naturally to the original motion.
Day-to-Night Relighting
Reference to VideoPrompt
Convert daylight to believable night while preserving action and geometry. Rebuild practical lights, shadows, reflections, sky exposure, and color temperature.
Window View Replacement
Reference to VideoPrompt
Replace only the exterior view. Match perspective, glass reflection, exposure, depth of field, and camera movement while fixing the interior.
Dialogue & Voice Editing
Replace spoken content and adjust the performance to match.
Dialogue & Performance Replacement
Reference to VideoPrompt
Replace the line with the supplied audio and adjust expression and delivery only as much as needed, preserving the environment and edit.
Precise Instruction Adherence
Apply localized edits while protecting everything not named.
Multi-Element Scene Editing
Reference to VideoPrompt
List every prop and visual substitution separately, then state that all unnamed composition, identity, timing, and motion must remain fixed.
Product, Sign & Dialogue Replacement
Reference to VideoPrompt
Replace the opening product, illuminated sign, final bag contents, and closing dialogue while preserving actors, camera, and event order.
Precision Costume Swap
First & Last FramePrompt
Swap only two magicians’ suit colors through smoke, keep glove colors fixed, close the curtain after the bow, then change its color.
Hand-Drawn Romance VFX
Reference to VideoPrompt
Add warm drawn marks around the couple, increase their brightness as they approach, and introduce pink brushwork only at the kiss.
The Three Input Modes
Text to Video
Prompt only. Use it when the look, action, camera, and sound can be described without a media reference.
First & Last Frame
Supply an exact opening image or an opening and closing pair, then describe how the shot moves between them.
Reference to Video
Combine up to nine images, three videos, and three audio clips, with no more than twelve files total.
Prompting MiniMax H3: Key Techniques
1.Assign a job to every reference
Say which file controls identity, wardrobe, location, motion, edit rhythm, voice, or sound.
2.Write a timed shot list
Use short time ranges for framing, action, transition, and sound whenever the clip contains more than one beat.
3.Direct the audio
Describe dialogue, ambience, music, foley, intensity, timing, and the exact moment an impact should land.
4.State what you do not want
Name likely failure modes such as soft dissolves, malformed text, black frames, visible seams, or the wrong genre.
5.Lock identity explicitly
List the face, hair, clothing, material, silhouette, type, or product details that must remain stable.
6.Pair every edit with a constraint
Describe the change and immediately state which neighboring elements must stay untouched.
7.Use camera and film language
Specify lens feel, framing, movement, focus behavior, exposure, grain, highlights, and shadow character.
8.Describe transitions as events
Explain the physical motion, cut point, blur, exposure change, and how the next shot settles.
Ready to build?
Start with one short shot and one clearly assigned reference. Test composition, event order, identity, motion, and sound before adding complexity.
Generation checklist
- Subject and setting are explicit
- Actions appear in playback order
- Every reference has one job
- Camera behavior is described
- Sound includes timing and texture
- Constraints protect fixed details