minimax hailuo h3 video generator ai
First frame
Last frame
Reference
Write a prompt...
Video

MiniMax H3 AI Video Generator

One prompt, a whole sequence. MiniMax H3 turns a script or an image into several shots in a single generation, cut together with clean transitions and synced sound. Hold the same character across every shot, in 2K, from 5 to 15 seconds, in six aspect ratios.

Founded in 2022
11M+ users
Commercial rights

Access 48+ leading AI models. Pay for one subscription.

Build a whole sequence in one generation

MiniMax H3 does not stop at one shot. Describe a sequence and it returns several shots together, five three-second cuts if you want, with transitions and pacing that follow your script.

A multi-shot scene comes back as one finished clip, not separate renders you stitch by hand. Longer sequences and fast action are harder to hold, so keep each shot's direction clear and generate a couple of options.

From script to a cut sequence in 3 steps

Nothing to install, no shots to stitch, no editing suite.

1. Open the Create video tool

Sign in, then open the Create video tool and select MiniMax H3 from the model menu in the prompt box.

2. Describe the shots

List each shot: the action, the camera, and any dialogue or sound. Our video prompt guide and the example below show how to structure a sequence.

3. Set the specs and generate

Pick an aspect ratio, set the length from 5 to 15 seconds, then press the arrow.

MiniMax H3 Specs

What you can set for each MiniMax H3 generation on getimg.ai. Sound comes with every clip, generated in the same pass as the picture.

Resolution 720p for quick drafts, 2K for the final cut.
Length Any whole number of seconds from 5 to 15.
Aspect ratios 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16.
Reference images Up to nine, for faces, outfits, products and style.
Video and audio references Up to three of each, with video clips adding up to 15 seconds.
First and last frame Pin the opening shot, the closing shot, or both.

Dialogue, effects, and score in the same pass

MiniMax H3 generates sound with the picture, not in a later step. Ask for a spoken line, an effect on a beat, or a music bed under the cut, and it comes back matched to the action across every shot. That drops the separate sound pass, so social and ad work reaches a finished cut sooner.

Reference consistency

Keep the same characters and props across every shot

Give MiniMax H3 up to nine reference images and it carries the same face, wardrobe, and product from one shot to the next. A sequence that used to drift between cuts stays on model, so a brand film or a recurring character holds together, no reshoot required.

Video and audio references

Hand it a clip to match, or a track to move to

References in MiniMax H3 are no longer limited to stills. Attach a video file and the model reads the motion in it. Attach an audio file and it has a sound to work against, rather than inventing one from the prompt alone.

Up to three video references and three audio references per generation, mixable with your reference images in the same shot. Video references can add up to 15 seconds in total across the three clips, and each audio file can run up to 15 seconds, so a full-length music track needs trimming first. MiniMax H3 Max has the same video limit.

Still photograph of a woman standing motionless on concrete steps in warm sunlight, attached as the image reference for an audio-driven dance generation.

Per-shot cinematography

Call the lens, the move, and the timing shot by shot

Set a focal length of 35, 50, 65, or 100mm, a camera move, and a duration for each shot. MiniMax H3 follows the direction shot by shot, so the cut arrives with the framing and pacing you asked for, not one fixed look. You block the scene the way a director of photography would.

How to Prompt MiniMax H3

MiniMax H3 reads your prompt as a shot list. The clearer each shot is, the closer the finished cut follows your plan.

To keep a person or product the same, attach reference images and call them Image 1, Image 2 and so on in the prompt. Short prompts work too, because prompt enhancement fills in the detail.

Number the shots with timings Give each shot its own line and a time range, like [0-3s], so every cut lands where you want it.
Give each shot a framing and a move Say wide, close-up or over the shoulder, then add a lens and a move: slow push-in, handheld or static.
Write the sound into the shot Put dialogue in quotes and place effects and music where they happen, like a door slam at 4 seconds.

[0-3s] Extreme close-up of an old watchmaker's hands opening the back of a gold wristwatch under a magnifying lamp, 100mm macro, static. [3-6s] Medium shot across the counter of a tiny Hong Kong watch repair stall at night, neon reflections in the wet street behind: a young customer with bleached curly hair and silver rings leans in and asks, "Can you save it?" 50mm lens, slow push-in. [6-9s] Reverse shot over the customer's shoulder: the watchmaker, a man in his seventies with round glasses and a grey cardigan, looks up and says, "Give me ten minutes." [9-12s] Wide shot from across the street as a green double-decker tram passes in front of the stall, handheld. Sound: rain on the awning throughout, soft mechanical ticking, a tram bell at 10 seconds.

Video
16:9
5s

MiniMax H3 Prompt Templates

Copy a template, fill in the brackets and generate. Each one is built for a different kind of sequence.

Product ad

[0-3s] Close-up of the [product] from Image 1 on [surface], slow push-in, [light]. [3-6s] Wide shot: [person] picks it up in [setting], handheld. [6-9s] Detail shot of [feature], rack focus. [9-12s] Hero shot of the product with "[TAGLINE]" in clean type. Sound: [music style], a soft [effect] on each cut.

Attach a photo of the real product as Image 1 so the shape and label stay the same in every shot.

MiniMax H3 vs MiniMax H3 Max

We gave all three models the same 6-second prompt: a bike courier racing through a Mexico City market, splashing through a puddle, then stopping at a taco stand, with three timed shots and sound cues. Each ran at its highest resolution in the Create video generator.

Explore More AI Features & Models

Frequently Asked Questions

Write the scene. Get the cut.

Describe your shots, set the lens and the length, and MiniMax H3 returns a cut sequence with sound. One generation, not a timeline to assemble.