Verified H3 prompt library

Watch the proof. Then copy the prompt.

6 examples, each pairing a MiniMax-published H3 output with the exact prompt shown beside it. No prompt-only filler and no videos borrowed from another model.

Explore verified examples Learn H3 prompting

Prompt ↔ output

Six verified examples. Zero guesswork.

These are deliberately few. Each output and prompt comes from MiniMax's official H3 feature gallery, and reference-dependent examples say so plainly.

Open the official MiniMax gallery

Official H3 output

2K · 15s · native stereo audioPrompt and output published together

01Visual creativeText to video

A glowing sketch comes alive in a real kitchen

Why it works

It directs texture, camera imperfections and a restrained sound palette, so the result feels captured rather than polished.

Prompt published with this output

369 characters

15s, 16:9. Live-action footage of a small kitchen at dusk fused with hand-drawn glowing animation. Old wooden table, half-washed mug, foggy jar, hanging dishcloth. Subtle handheld shake, hesitant close-focus, backlit exposure fluctuation — as if someone at home rushed to capture something inexplicable. Sound: kitchen ambience plus the creature's soft electronic hums.

Text-only generation — this opens Text to video with no reference assets required.

Verify this pair on MiniMax

Official H3 output

2K · 15s · native stereo audioPrompt and output published together

02Brand filmReference guided

A fashion film built from four precise references

Why it works

Every reference has one job—mood, character, product or logo—while the story stays simple enough to resolve in 15 seconds.

Prompt published with this output

335 characters

A 16:9 high-fashion brand campaign. Image 1: mood, setting, film grain. Image 2: character. Image 3: bag. Image 4: ending logo. Core story: on a desert highway beside a vintage car, the woman takes a black bag from the trunk, shares a brief quiet moment with the man, then leaves. Tone premium and cool; edit agile and fashion-forward.

This official output uses 4 reference images, so it opens Reference to video. Add matching assets before expecting a comparable result.

Verify this pair on MiniMax

Official H3 output

2K · 15s · native stereo audioPrompt and output published together

03NarrativeReference guided

A vertical vampire-romance trailer with locked leads

Why it works

It locks the protagonists and location, then spends the shot budget on close framing, eye contact and relationship tension.

Prompt published with this output

442 characters

15s, 9:16 vertical trailer for an overseas live-action vampire-romance short drama. Image 1: male and female leads. Image 2: setting. Story: an innocent human heroine wanders into a forbidden castle and wakes the sleeping vampire aristocrat; he develops a dangerous, controlling fascination — she fears him but doesn't submit. ReelShort / DramaBox feel: dark romance, danger, destiny. Medium-close and close shots for eye contact and tension.

This official output uses 2 reference images, so it opens Reference to video. Add matching assets before expecting a comparable result.

Verify this pair on MiniMax

Official H3 output

2K · 15s · native stereo audioPrompt and output published together

04ProductReference guided

A product ad that specifies the design, not just the mood

Why it works

Concrete product geometry gives H3 details to preserve; the attitude and white-cyc direction keep every shot in one campaign world.

Prompt published with this output

295 characters

9:16 high-fashion eyewear commercial on a minimal white cyc. Image 1: two full-body female models. Image 2: facial details. Image 3: eyewear design — wrap-around curves, sharp cat-eye/goggle hybrid silhouettes, mirror reflections, streamlined temples. Keep the runway attitude and cool demeanor.

This official output uses 3 reference images, so it opens Reference to video. Add matching assets before expecting a comparable result.

Verify this pair on MiniMax

Official H3 output

2K · 15s · native stereo audioPrompt and output published together

05UI motionReference guided

A landing page turned into a motion brief

Why it works

The prompt defines hierarchy, typography, materials, pacing and interaction states instead of asking vaguely for a “cool website.”

Prompt published with this output

329 characters

Product landing page UI/UX motion demo — smooth vertical scroll, hero subject is product image 1. Bold, tilted, oversized sans-serif typography. Background weaves speed-oriented motion light with dark carbon-fiber and sport-mesh textures. Tight, high-impact rhythm; strong hover interactions like visual zoom and color inversion.

This official output uses 1 reference images, so it opens Reference to video. Add matching assets before expecting a comparable result.

Verify this pair on MiniMax

Official H3 output

2K · 15s · native stereo audioPrompt and output published together

06Game creativeReference guided

A first-person action loop with readable recoil

Why it works

Viewpoint, handling, movement, light and recoil are described as one causal action loop, keeping the camera language game-authentic.

Prompt published with this output

419 characters

First-person, eye-level, handheld gameplay camera. Simulates a player in a modern-warfare FPS — assault rifle held in both hands, advancing along a military base perimeter, crosshair sweeping the passageway, firing a few rounds at a distant target, then pressing forward. Cold natural light meets smoke and muzzle flashes; photorealistic, 3A-game weapon and dust texture. Subtle handheld shake with recoil kick on fire.

This official output uses 1 reference images, so it opens Reference to video. Add matching assets before expecting a comparable result.

Verify this pair on MiniMax

MiniMax H3 prompt tutorial

Write a timeline, not a pile of adjectives.

This tutorial distills MiniMax's official H3 writing guide into a practical workflow you can use before every render.

Read MiniMax's official base guide Read the full MiniMax H3 prompt guide

The directing loop

Four decisions before you press Generate

Choose the generation mode first

Text-to-video invents the complete timeline. Image-to-video must develop forward from the first frame. First-and-last-frame mode must describe the continuous path between both anchors. Full-reference mode assigns a specific job to every image, video and audio file.

Describe change in playback order

Open with style and composition, then state what the subject does, how the camera responds and what visibly changes. A later shot should start with a real cut time such as [Shot 2] At 00:05.000 — not a loose list of scenes.

Direct one camera move at a time

Use motion type, amplitude and speed when they matter: “the camera pushes in with small amplitude at slow speed.” Prefer camera movement over a cut when only framing or angle needs to change.

Write audio as part of the scene

Keep dialogue with the speaker and shot. Put ambience, footsteps, impacts and non-verbal sounds in overall_soundscape. Put audience-only score in non_diegetic_music, or write N/A when there is none.

The official three-part body

Give picture, sound and score separate jobs

MiniMax's base guide uses three fields. This separation keeps visible action, physical sound and background music from competing with each other.

integrated_multimodal_description

The shot-by-shot timeline: visual style, initial composition, subject action, camera motion, cuts, dialogue and synchronized events.

overall_soundscape

One compact paragraph for ambience, physical action sounds and non-verbal human sounds across the full video.

non_diegetic_music

Audience-only music described through instruments, tempo and dynamics. Use N/A instead of inventing a score you do not want.

Fill-in template

H3 text-to-video starting structure

integrated_multimodal_description: [Shot 1] [Style], [framing] establishes [subject] in [setting]. The subject [action and visible change]. The camera [motion type + optional amplitude + speed]. [Shot 2] At 00:05.000, the camera cuts to [new information, reaction or result].

overall_soundscape: [Ambience]. [Physical action sounds]. [Non-verbal human sounds, if any].

non_diegetic_music: [Instruments + tempo + dynamic change], or N/A.

Camera words that do different jobs

H3's official vocabulary separates lens changes from physical camera movement. That distinction makes framing more predictable.

Push in / Zoom in

Push in moves the camera forward and changes perspective. Zoom in changes focal length while the camera body stays still.

Pan / Truck

Pan pivots the lens left or right from one position. Truck moves the entire camera sideways through space.

Tracking shot

Use it when the camera should follow a moving subject and preserve their importance in frame.

Static shot

State it when both the camera position and lens must remain still; avoid later wording that implies reframing.

Change the structure with the mode

Text to video

Build the complete audiovisual timeline from text. Start Shot 1 directly with style, framing and action.

Image to video

Treat the uploaded image as the exact first-frame anchor. Preserve identity, clothing, colors, objects and spatial relationships, then describe forward motion.

First + last frame

Usually favor one continuous shot. Describe observable intermediate changes that progressively close the gap between the opening and ending frames.

Full reference

Name what each asset contributes—identity, setting, action, camera movement, voice or soundtrack. Do not expect H3 to guess why a file was uploaded.

Open MiniMax's full-reference guide

Pre-render check

Five things to verify

Every shot contains an observable action or state change, not only a visual description.
Every cut reveals new information; later shots use increasing timestamps inside the clip duration.
Every reference asset has one explicit role and the details that must remain fixed are named.
Dialogue keeps the exact wording and language, with a stable speaker across shots.
Ambience and physical sounds are separate from audience-only background music.

What to know before you copy

What makes a prompt H3-specific?

H3 accepts a unified audiovisual direction: shot order, camera movement, reference roles, dialogue, physical sound and music. The official format is more like a compact directing brief than a still-image description.

Do I have to use the full structured format?

No. The official gallery shows that concise natural-language prompts can work. Use the three-part structure when timing, cuts, dialogue or audio control matter; keep simple ideas simple.

How do you know these videos are MiniMax H3 outputs?

MiniMax publishes every prompt and playable output together on its official H3 Feature Highlights page. Each example above links directly to that primary source; the page does not include unverified community clips or other models.

Will the copied prompt reproduce the same video?

Not exactly. Generation is variable, and five examples also use published reference images. Each card states the reference count so you can distinguish a text-only prompt from a reference-guided result.

Direct one clear shot before you direct ten.

Start with the text-only kitchen example or the H3 template, render once, then change one variable at a time.

Open the H3 generator