9 MiniMax H3 skills: which workflow should you use?

The nine MiniMax H3 skills in the official repository are not nine interchangeable prompt presets. One translates a multimodal brief into H3's structured prompt formats. The other eight organize a particular kind of production, from product advertising and lyric-driven music videos to papercraft explainers and two-player game intros.

That distinction is the fastest way to choose well. Use h3-prompt-writing when the deliverable is a technically structured H3 prompt. Choose a style or workflow skill when the deliverable includes creative direction, source-asset checks, approval gates, storyboards, generation, or assembly. This guide maps all nine packages documented in the official MiniMax H3 skills directory, with emphasis on where each workflow starts and stops.

The nine MiniMax H3 skills at a glance

The table is a routing map, not a quality ranking. Several skills can touch prompts, storyboards, video, and audio, but each protects a different creative contract.

1. h3-prompt-writing: the prompt-structure specialist

Choose h3-prompt-writing when you already know the scene and need to express it in the structure H3 expects. It supports five modes: text-to-audio-video (T2VA), first-frame-to-audio-video (I2VA), first-and-last-frame-to-audio-video (FL2VA), last-frame-to-audio-video (L2VA), and full-reference Ref2VA.

For the four base or keyframe modes, the skill follows one reference guide and emits integrated_multimodal_description, overall_soundscape, and non_diegetic_music in that order. Ref2VA uses a separate six-section guide so image, video, and audio references can be labeled and assigned explicit jobs.

Use this as the finishing layer when the creative decisions are already locked. If you still need to discover a story, verify brand facts, design a visual language, or approve frames, start with one of the production skills below and use prompt structuring after the plan is stable.

2. minimalist-product-ad-generator: product-first advertising

This skill starts with product images or related assets, then checks variants, extracts defensible selling points, and confirms the production brief. Its control system uses three independent product anchor photos plus a precise beat-and-text storyboard rather than a single multi-panel reference sheet.

Choose it when the product itself must remain the visual anchor in an e-commerce short or launch film. The workflow also handles concise English ad copy, camera direction, generation, technical BGM, beat-synced editing, and delivery review. It deliberately keeps music out of the initial intake and resolves it later in the workflow.

The key boundary is product focus. A founder speaking to camera, a complex software walkthrough, or an open-ended editing job should not be forced through this pipeline.

3. 3d-animation-short-generator: story-first animated shorts

Use this skill for a complete stylized 3D narrative, not merely a 3D look. It moves in production order through a project brief, story outline, character and environment cards, a standardized six-column shot table, text storyboards, model selection, single-shot generation, full assembly, BGM matching, and final review.

Its gates are part of the value. Screen size and total duration are settled early, and creative choices are confirmed before expensive downstream work. The skill names H3 as the default video model and Seedance 2.0 as a fallback, so it is a broader directing workflow rather than an H3-only prompt template.

Choose it when character consistency, scene continuity, performance, camera direction, and timing have to survive across shots. For a one-off clip or prompt consultation, it is more process than the job needs.

4. papercraft-stop-motion-explainer: layered knowledge design

This is the deeper of the two paper-based explainer workflows. It turns a science, education, or general-knowledge topic into a tactile miniature world with paper characters, layered diorama sets, props, physical shadows, camera rules, transitions, narration, and review criteria.

The complete route can cover creative directions, four to seven visual planes, asset planning, preview images, image-series prompts, short video prompts, storyboards, editing rhythm, voiceover cleanup, negative prompts, and quality review. It also defines lightweight bypasses when the user asks for only one asset, such as a still prompt or five-second image-to-video prompt. H3 is the default when the workflow advances into video generation, unless the user selects another available model or H3 is unavailable.

Choose it when the explanation itself benefits from mechanisms, cross-sections, pop-up staging, or a paper narrator. If the source is a short opinion or narration beat and the goal is editorial B-roll, the next skill is usually a tighter fit.

5. brand-promo-video-generator: provenance before promotion

This workflow is for a brand, product, website, app, shop, or personal project with verifiable source material. It can start from logos, product images, interface screenshots, official links, or a clear idea, but it builds a brand truth sheet and provenance manifest before it chooses a story spine.

That makes it the right choice when accurate identity and claims matter as much as polish. The workflow plans exact beats and motion, requires hard confirmation before generation, creates the necessary image, video, voice, or music assets, and performs a pre-delivery review. Its source policy rejects identity assets from logo aggregators, thumbnails, fan recreations, and AI-generated substitutes.

Use the minimalist product-ad skill when a physical product and its variants are the center of the film. Use this one when the wider brand story, sourced interface, capability claims, use cases, and call to action must stay coordinated.

6. music-video-subtitle-generator: lyrics as scene direction

The installable skill name is music-video-subtitle-generator, although its repository folder is mv-subtitle-skill-confirmed. It is built for music videos and emotional short films where rhythm, lyrics, performance, references, camera language, and on-screen typography form one system.

The workflow locks format, duration, music window, lyrics, and reference roles before developing the creative contract. It can split work longer than a single generation into connected shots, coordinate audio continuity, design beat-reactive spatial typography, audit prompts, and provide stitching guidance. H3 is one supported generation route, not the only one named by the skill.

Choose it when text is composed as part of the picture. Ordinary subtitles, caption correction, and generic video edits belong in an editing workflow because this skill treats lyrics and typography as art direction.

7. co-op-game-intro-generator: a narrow, useful game-menu recipe

This skill has the most specific content contract: two players, a game title, a visual style, and optional character references. It first creates a confirmation image from a fixed menu framework, including coordinated palette, UI, characters, typography, buttons, and icons.

Only after approval does it rebuild the final H3 video prompt from the accepted image, final UI copy, player and game data, event timing, motion instructions, and negative constraints. Two bundled prompt templates are mandatory; the skill says to stop rather than improvise if either is missing.

Use it for a character-led game menu, opening animation, concept reel, or social clip. It does not build a playable game, a complex application interface, or a generic title sequence with no player-card structure.

8. paper-collage-explainer-generator: editorial metaphor and B-roll

This skill turns a narration line, viewpoint, story beat, or abstract topic into premium halftone paper collage. Its visual system combines bold color fields, black-and-white photographic cutouts, cardstock accents, warm keylines, paper shadows, and stop-motion assembly.

The workflow first seeks approval for a production plan and storyboard, then builds still-frame specifications and asks for still approval before generating clips. Its default audio rule is unusually clear: preserve tactile collage sound effects, but do not add BGM, voiceover, or subtitles unless the user requests them.

Choose it when you need concise visual metaphors, editorial explainers, or social B-roll. Choose papercraft instead when the subject needs characters, layered mechanisms, a diorama, or a fuller educational production package.

9. handdrawn-live-video-generator: one controlled hybrid shot

This workflow is narrow by design. It writes a 15-second, 16:9 prompt in the user's dominant language for a live-action space containing a rough, luminous hand-drawn entity. In the first three seconds, that drawing must visibly contact a real hand or object; it then continuously morphs, escapes, and is followed by a handheld camera that reacts slightly late.

The material language is crayon, chalk, colored pencil, pastel, or rough brush—not clean vector animation, smooth neon, 3D CG, plush texture, or horror. The skill waits for explicit confirmation before recommending H3 generation and reviews contact realism, morph continuity, camera delay, stroke texture, and tone.

Use it for one surreal, physically legible fusion shot. A multi-scene story, a polished animation pipeline, or a horror transformation breaks the skill's core structure.

A MiniMax H3 skill decision tree

Start with the deliverable, then narrow by the asset and approval requirements:

1. You only need an H3-ready prompt. Use h3-prompt-writing.
2. You need a complete production workflow. Continue by content type:
   • Physical product at the center → minimalist-product-ad-generator
   • Wider brand, app, website, or shop story → brand-promo-video-generator
   • Multi-shot stylized 3D narrative → 3d-animation-short-generator
   • Lyric- and rhythm-led video → music-video-subtitle-generator
   • Two-player game menu or intro → co-op-game-intro-generator
   • Single live-action and hand-drawn hybrid → handdrawn-live-video-generator
   • Explainer or abstract concept → choose between the two paper workflows:
     • Layered educational diorama and mechanisms → papercraft-stop-motion-explainer
     • Editorial metaphor, narration beat, or collage B-roll → paper-collage-explainer-generator

If two routes still appear valid, choose the one whose required inputs you already possess. Product photographs favor the minimalist ad workflow; authorized logos, official links, and interface captures favor brand promo; locked lyrics and a music window favor the MV workflow. A skill's intake gate usually reveals its intended job more reliably than its visual style name.

How to install the skills

The repository uses the skills CLI. To inspect the available packages before installing anything, run:

npx skills add https://github.com/MiniMax-AI/MiniMax-H3 --list

Install all nine with the repository's documented command:

npx skills add https://github.com/MiniMax-AI/MiniMax-H3 --skill '*'

For a smaller agent setup, install only the workflow you need:

npx skills add https://github.com/MiniMax-AI/MiniMax-H3 --skill h3-prompt-writing

Replace h3-prompt-writing with the installable skill name from the table. Installing one skill first is the clearest way to learn its approval gates and output contract without creating competing instructions.

Bring the selected workflow into OmniArt

The right skill should reduce ambiguity before generation: it should leave you with a locked brief, clearly assigned references, an approved visual direction, or a structured H3 prompt. It does not guarantee that every generation will preserve a face, product, line of text, physical interaction, or timing cue. Keep approval frames and source assets, review the full clip, and regenerate or edit when a critical constraint drifts.

Open MiniMax H3 on OmniArt when your selected workflow reaches the generation handoff. Keep image, video, audio, and prompt decisions together in the same creative workspace, then continue into OmniArt's other image, video, audio, or music tools when the production needs another modality. For a deeper look at reference labels and timeline structure, use the MiniMax H3 prompt guide.

The practical rule is simple: choose by workflow boundary, not by aesthetic resemblance. A narrow skill with the right inputs and checkpoints is more useful than a broader-looking template that owns the wrong production problem.