Video
Image
Audio
More
Discord

Write a video brief covering subject, action, camera, lighting, and sound — then compare your result with six real MiniMax H3 outputs below.

Choose a complete shot, then replace the subject, setting, action, dialogue, and sound before generating.

Sign in to upload

Sign in to upload

*
Start with

0/2,000

*
*
515
70credits
Example Video
MiniMax H3 — First batch at dawn thumbnailMiniMax H3 — Rooftop pursuit thumbnailMiniMax H3 — Paper becomes a swan thumbnailMiniMax H3 — Arena speed climb thumbnailMiniMax H3 — Ten-metre dive thumbnailMiniMax H3 — Floor routine thumbnail
Real AI-generated video
Sound in every clip
Six different motion tests
Review notes included

SHOT / SOUND / RESOLVE

You can direct the entire clip, not just the first frame.

This bakery example shows how a single prompt controls the whole sequence: it sets the location, drives the action, delivers a line of dialogue, cuts to a detail shot, and ends on the product. The sound follows the same structure beat by beat.

Example 01 / 061920 × 1080 · 10.1s

MiniMax H3 / Dialogue + ambience

First batch at dawn

Dialogue, a motivated cut, room tone, and a musical finish meet in one MiniMax H3 bakery scene.

Exact generation prompt

integrated_multimodal_description: [Shot 1] Live-action, cinematic, a medium-wide shot frames a baker opening the shutters of a small street bakery before sunrise. The camera pushes in with small amplitude at slow speed as the middle-aged baker with a calm, slightly raspy voice (…

What to inspect

Check the voice-to-cut handoff and whether the loaf stays recognizable across the shot change.

  • two shots
  • spoken line
  • stereo sound
00:00

Establish

Predawn bakery, medium-wide frame.

00:02

Act

Shutters open; the loaf lands.

00:04

Speak

Quoted line, named voice, physical sound.

00:05

Cut

Dialogue carries into the detail.

00:10

Resolve

Product holds; music fades.

PROMPT LAB

How to write a prompt that actually works.

The best prompts read like a shot list: every sentence changes something you can see or hear. Start from the opening frame, build the action in order, and always describe the ending — otherwise the model decides for you.

  1. 01
    Frame

    Set the opening shot

    Start with what the camera sees: the subject, the location, the framing, and the lighting. Everything that follows will build on this first image, so make it specific.

  2. 02
    Action

    Describe what happens, in order

    Write the action as a sequence of concrete verbs — shutters open, the loaf lands on the counter, steam rises. Each motion should cause the next so the clip reads as one continuous event.

  3. 03
    Camera

    Give the camera one job

    Pick a single camera move — a slow push, a tracking shot, a locked frame, an orbit — and state its speed. Stacking multiple moves almost always produces drift.

  4. 04
    Sound

    Write sound next to its cause

    If a character speaks, quote the line and name the speaker. Then layer in the ambience (street noise, kitchen hum) and any music. Place each sound cue right beside the action that produces it.

  5. 05
    Resolve

    End on a frame you can use

    Describe the final image explicitly: the product angle, the character's expression, the landing pose, or where the loop should restart. Without this, the clip drifts to a random ending.

Four quick MiniMax H3 prompt checks

  • One continuous action always looks better than five disconnected ideas crammed into one clip.
  • Only add timestamps where the scene actually changes — a beat, a cut, or a new sound cue.
  • If a face, logo, or shape must stay consistent, say so explicitly in the prompt.
  • Watch the result twice: once for picture, once with sound on. They fail in different ways.
Example 02 / 061920 × 1080 · 10.1s

MiniMax H3 / FPV camera motion

Rooftop pursuit

This fast MiniMax H3 shot asks the camera to bank with a runner instead of observing from a fixed cinematic position.

Exact generation prompt

A skilled parkour athlete sprints through the city, leaping and doing many flips over urban obstacles. FPV tracking shot, swoops and banks with him.

What to inspect

Track body shape through each landing and judge whether the banking camera keeps the action readable.

  • one take
  • fast motion
  • camera banking

MULTIMODAL CONTEXT

You don't have to start from scratch.

MiniMax H3 accepts images and audio as starting context — use them to lock down a character's look, a specific composition, a voice, or a destination frame. The generator shows which input types are available and their limits.

Example 03 / 061920 × 1080 · 10.1s

MiniMax H3 / Object transformation

Paper becomes a swan

A minimal prompt gives MiniMax H3 one material, one transformation, and one final object, making the motion easy to judge frame by frame.

Exact generation prompt

A white paper beautifully and elegantly folds itself into an origami swan

What to inspect

Check causal folds, stable paper texture, and whether the final silhouette reads clearly as a swan.

  • simple prompt
  • material continuity
  • clear ending
01

Prompt only

Just write a text prompt and let the model invent the scene. Best when the action, camera work, and sound matter more than matching a specific visual.

TEXT → NEW VIDEO
02

Start from an image

Upload an image when you already have the subject or composition you want. In the prompt, be clear about what should move and what must stay exactly as it is.

IMAGE → DIRECTED MOTION
03

Add references

Use reference images or audio to communicate a character's appearance, a visual style, a specific movement, or the voice and sound you want in the clip.

CONTEXT → CONSISTENT SHOT

According to the MiniMax H3 model card, the model supports 4–15 second clips at 24 fps with stereo audio and up to 2K output resolution. The generator above exposes the available controls.

12 SHOTS TO TRY

Twelve shots worth trying first.

Each idea below is small enough to fit a single clip but specific enough to test something real — motion, continuity, sound, or final-frame quality. Pick one, write the prompt, and see what you get.

01

Product reveal

A single camera move around the product with stable geometry and moving reflections, ending on a clean packshot.

02

Short dialogue

One speaker delivers one quoted line while doing something natural — pouring coffee, looking up from a desk.

03

Material change

Show a material becoming a finished object — clay on the wheel, dough in the oven — in connected stages.

04

Food close-up

Steam, texture, one slow lens push, kitchen sounds in the background, and a final frame that looks appetizing.

05

Athletic action

A run, a jump, a landing, and the reaction — keep the contact points readable so the motion feels physical.

06

Character entrance

Reveal a character through posture, wardrobe, lighting, and footsteps before they say or do anything.

Example 04 / 061920 × 1080 · 10.1s

MiniMax H3 / Broadcast-style action

Arena speed climb

A short prompt tests whether MiniMax H3 can invent the venue, broadcast framing, athletic motion, and crowd energy without a long art-direction brief.

Exact generation prompt

a dog speed climbs up a climbing wall at the olympics

What to inspect

Inspect paw contact, wall geometry, upward progress, camera continuity, and crowd sound.

  • sports framing
  • animal motion
  • crowd sound
07

Vertical social hook

Something eye-catching in the first second, subject centered for mobile, and a clean loop point at the end.

08

Environment reveal

Start on a tactile detail — rain on a leaf, rust on a railing — then pull back to reveal the full scale of the place.

09

Quiet performance

A face that shifts from one expression to another with consistent eyeline and quiet ambient sound. No dialogue needed.

10

Previsualization

Test your blocking, lens choice, timing, and composition before committing to a live shoot.

11

Music-led beat

Set the tempo in the prompt, align the action to the beat, and end on a musical accent.

12

Illustration in motion

Animate small details — flickering candles, turning pages — while keeping the linework, palette, and layout intact.

EXPORT GATE

A good still frame doesn't mean the video works.

Video has to hold up across every second, not just one screenshot. Run through these checks before you download, share, or iterate on the next version.

Example 05 / 061920 × 1080 · 10.1s

MiniMax H3 / Multi-stage motion

Ten-metre dive

MiniMax H3 has to preserve one subject through takeoff, rotation, descent, and entry while maintaining the visual grammar of televised sport.

Exact generation prompt

televised footage of a cat is doing an acrobatic dive into a swimming pool at the olympics, from a 10m high diving board, flips and spins

What to inspect

Follow one subject from platform to water; every phase should belong to the same dive.

  • action phases
  • wide to close
  • subject tracking
Identity

Does the subject stay recognizable?

Scrub through the clip and compare the face, silhouette, wardrobe, and any distinctive details from start to finish.

Physics

Do contact and weight feel right?

Watch for feet that float above the ground, hands that pass through objects, shadows that point the wrong way, and fabric that ignores gravity.

Camera

Can you describe the camera move in one sentence?

A good result has a camera that moves with purpose and stops cleanly. If it drifts without direction, the prompt was too vague.

Audio

Does every sound match what you see?

Check that dialogue is in sync with lip movement, that effects line up with the action, and that the music level doesn't bury the important sounds.

Example 06 / 061920 × 1080 · 10.1s

MiniMax H3 / Run, flip, land

Floor routine

This MiniMax H3 output turns a compact sequence of verbs into a beginning, peak action, landing, and crowd response.

Exact generation prompt

a cat is doing a gymnastic routine on the mat, it runs and jumps, does a flip, then lands perfectly, to standing applause

What to inspect

Check action order, landing pose, mat contact, and whether applause arrives after the finish.

  • ordered action
  • timed reaction
  • sound cue

MINIMAX H3 FAQ

Questions creators ask before the first run.

What is MiniMax H3?

MiniMax H3 is a multimodal AI video model that generates short video clips with synchronized sound — dialogue, effects, ambience, and music all in one pass. On Voor, you can write a text prompt and optionally attach up to five reference images and one audio reference. Video-reference uploads are not supported yet.

What can I make with MiniMax H3?

Cinematic concepts, product shots, social media clips, material transformations, previsualization for live shoots, character moments, and short ads. The key is to give each generation one clear action and a defined ending so you can judge the result.

Does MiniMax H3 generate audio with the video?

Yes — MiniMax H3 generates synchronized stereo audio alongside the video. It can produce dialogue, ambient sound, sound effects, and music. To get the best results, name the speaker, quote their line, and describe each sound next to the action that causes it. Every example on this page includes audio.

How long can a MiniMax H3 video be?

The generator supports clips from 5 to 15 seconds, adjustable in one-second steps. In practice, around 10 seconds tends to work best — long enough for one continuous action or two clearly timed beats, short enough to stay coherent.

What resolution does MiniMax H3 support?

The Voor generator doesn't currently expose a resolution selector — the output size is determined upstream. You can choose between 16:9 and 9:16 framing, and the delivered file will reflect the actual output dimensions.

Can MiniMax H3 animate an image?

Yes. You can upload up to five reference images to anchor the visual. For the best results, use clean compositions, mention each referenced subject by name in the prompt, and be explicit about what should move versus what needs to stay exactly as it is.

How do I write a good MiniMax H3 prompt?

Think of it as a mini shot list. Start with the subject and location, then describe the action in order, add camera direction, lighting, quoted dialogue, environmental sound, music, and the final frame. Use concrete verbs and write events in the order they should happen.

How should I review a MiniMax H3 result?

Play the full clip from start to finish — don't just check a thumbnail. Watch for identity consistency, realistic contact and weight, a clean camera move, correct action order, dialogue lip sync, sound timing, and whether the final frame matches what you asked for.

Can I use a MiniMax H3 video commercially?

That depends on several factors: the rights to your prompt and reference materials, the model's terms of use, consent requirements, and your publishing context. Before releasing anything, carefully check the output for protected logos, recognizable faces, copyrighted music, and any unintended details.

YOUR NEXT SHOT

Turn one idea into motion and sound.

Pick a starting brief from the presets, swap in your own subject and action, check the credit estimate, and generate when you're ready. If the job can stay at 480P or 768P and you want to compare several directions quickly, try the speed-tuned MiniMax H3 Max route instead; keep this original H3 workflow for 2K and broader reference work. For a short prompt or a single approved first frame, compare MiniMax H3 Max Turbo with Balanced or Quality prompt expansion, 5–15 second clips, and native audio at 480P or 768P.

Open the generator

To steer upcoming scenes while watching a continuous stream, use H3 Max Director and send new directions during the session.