Open-weight video model by MiniMax

MiniMax H3 Video Generator

Direct scenes with text, frames, images, video, and audio references—then plan the whole sequence before you render.

Start from a shot brief

Resolution: Up to 2K
Duration: 5–15 seconds
Playback: 24 FPS
Audio: 32 kHz stereo

Describe the fruit, scene, action, camera, dialogue, and visual style...

38 credits

Enter to generate · Shift + Enter for a new line

0/5000

Story Workshop

Decide how you want AI to help

Choose a conversation starter, review the brief, then open it in your preferred AI chat.

Choose a starting prompt.

Prepared chat brief.

I want to create a love story starring stylized anthropomorphic fruit characters with clearly recognizable fruit forms and textures, along with expressive, human-like faces and bodies.

I want the story to explore a relationship that can feel romantic, awkward, funny, heartfelt, or emotionally complicated. Please develop a complete and compelling love story with a clear beginning, growing chemistry, escalating relationship tension, meaningful reversals, and an emotionally satisfying ending. Please also provide a concise character guide for each recurring character, covering their fruit identity, defining visual traits, personality, and a reusable base appearance suitable for generating character reference images. Keep each character’s core visual identity consistent throughout the story, while allowing clothing and accessories to change naturally with the setting and character development.

Once we have discussed and settled on the story, continue by breaking it into a sequence of video clip descriptions in story order. Each clip should have a short title followed by a compact natural-language description that can be used directly as the story prompt for generating one video clip.

Together, these clip prompts should form a continuous story. When designing them, keep most clips focused on one or two visible characters so the video model can understand them clearly. Each clip should center on one clear, continuous action or emotional beat within a single setting and continuous moment. Use most of the description to convey the story itself: what the characters do, how they react and interact, and how the moment changes or advances the plot. The characters’ visual identities will be carried by separately provided reference images, so recurring characters can be referred to by name. Include appearance, clothing, or accessory details when they matter to the specific moment.

1901/5000

The AI will guide the discussion and finish with one copy-ready video prompt.

Open in your preferred AI chat.

Open in ChatGPT Open in Claude

Your prepared prompt is sent to the service you choose and may appear in its URL, browser or account history, and logs. That provider's terms and privacy policy apply.

Resolution: Up to 2K
Duration: 5–15 seconds
Playback: 24 FPS
Audio: 32 kHz stereo

Watch the model work

Official H3 examples

See the range MiniMax chose to introduce H3: character detail, title design, interactive concepts, and vertical motion graphics.

View MiniMax H3 launch

MiniMax H3 official launch artwork — Official MiniMax H3 launch artwork

Motion and sound

A branded motion study with synchronized visual beats.

Film title sequence

Cinematic opening titles with controlled typography and mood.

Interactive concept

A polished interface concept presented as a moving product shot.

Vertical motion poster

A portrait-format animated poster built for social feeds.

Media is sourced from the model publisher's official showcase.

Why this model matters

Control more than the prompt

H3 accepts more than a text description. Its unified context can combine frames, images, video, and audio so direction lives in one brief.

Write sound into the scene

Direct dialogue, ambience, sound effects, and music alongside the picture with native stereo output.

Control the beginning and end

Use first and last frames to define where a shot begins and where its motion must land.

Mix reference types

Guide character, object, style, motion, or sound with image, video, and audio references in one context.

Aim for a detailed finish

Plan for 24 FPS output and an up-to-2K regeneration workflow when final detail matters.

Start with direction

MiniMax H3 base model prompt ideas

These starters are shaped around H3's multimodal strengths. Load one into the current tool, change the fruit, setting, sound, or camera, and make it yours.

Midnight confession

Dialogue, ambience, and two deliberate camera beats.

First-to-last reveal

A beginning and ending composition for controlled motion.

Directed performance

Character, movement, and camera direction in one brief.

Market launch spot

A compact three-shot sequence with readable scene text.

Same foundation, different priorities

MiniMax H3 vs H3 Max

Choose H3 when broad reference control and the 2K regeneration path matter most. Choose H3 Max when fast iteration and instruction following matter more than maximum resolution.

CompareMiniMax H3MiniMax H3 Max by fal

Best for

MiniMax H3

Omni-modal control and a higher-resolution finish

H3 Max by fal

Fast, prompt-faithful iteration

Resolution

MiniMax H3

Up to 2K through regeneration

H3 Max by fal

480p or 768p

Direction

MiniMax H3

Text, frames, images, video, and audio

H3 Max by fal

Text, start/end images, and references

Core strengths

MiniMax H3

Reference control, native stereo, and editing

H3 Max by fal

Speed, adherence, aesthetics, and visual coherence

Explore MiniMax H3 Max by fal

Prompt recipe

Write the whole scene

A strong video brief reads like compact direction, not a bag of style words. Give the model an ordered scene it can execute.

MiniMax H3 launch Official H3 repository H3 on fal

1Name the subject and the visible goal
2Order the action as timed beats
3Direct framing, lens, and camera movement
4Describe dialogue, ambience, effects, and music
5Finish with the look and constraints that must hold

Straight answers

MiniMax H3 base model FAQ

Who made MiniMax H3?

MiniMax released H3 as an open-weight omni-modal video model. fal is one of the platforms that serves the base model through an API.

What is the difference between H3 and H3 Max?

H3 is MiniMax's base model, with broad multimodal reference controls and an up-to-2K regeneration path. H3 Max is a separate fal Research post-trained variant focused on faster generation, prompt adherence, and aesthetics at 480p or 768p.

Can MiniMax H3 generate audio?

Yes. MiniMax documents native 32 kHz stereo audio, so a prompt can direct dialogue, ambience, sound effects, and music together with the picture.

What can I use as a reference with MiniMax H3?

H3 can combine text, first and last frames, images, video, and audio references in one context. MiniMax also documents editing and an up-to-2K regeneration workflow.