h3-prompt-writing Skill (MiniMax) for Claude: Write H3 Video Prompts

Makes your assistant write MiniMax H3 video prompts in the exact format H3 expects, for text, keyframe and reference inputs.

Installs: 12K; on skills.sh
Stars: 11.1K; on GitHub
Files: 3; besides SKILL.md
Updated: Oct 2026; last repo commit

h3-prompt-writing makes your assistant write or rewrite prompts for MiniMax H3, a video model that also generates sound. It handles five input modes: text only (T2VA), first frame (I2VA), first and last frame (FL2VA), last frame (L2VA) and full reference mode with images, video and audio (Ref2VA). It picks the mode, reads the matching guide in the skill, and keeps the exact field names and section order.

The output is a prompt, not a video. Base modes use three fields: integrated_multimodal_description, overall_soundscape and non_diegetic_music. Reference mode uses six sections and keeps labels such as <Picture 1> consistent. The skill's frontmatter says it needs no external API calls or MiniMax Hub tools.

Good for

People generating video with MiniMax H3 who want prompts in the right structure
Creators who start from a first or last frame and need the prompt to line up with it
Teams rewriting multi-image, video and audio references into one H3 prompt

Install h3-prompt-writing

Claude Code

Copy this prompt.

Install the h3-prompt-writing skill in Claude Code. Run this command:

npx skills add https://github.com/minimax-ai/minimax-h3 --skill h3-prompt-writing

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Write MiniMax H3 video generation prompts for T2VA, I2VA, FL2VA, L2VA, and Ref2VA. Use when rewriting multimodal requests into H3 prompt structures, composing integrated_multimodal_description, overall_soundscape, and non_diegetic_music, aligning keyframes, or defining reference labels for images, videos, and audio.

Sections in SKILL.md

Workflow
Base Modes
Full-Reference Mode
Output Rules
Tips for Better Results

Key facts

Author: MiniMax (MiniMax-AI/MiniMax-H3 repo)
License: MiniMax H3 Community License Agreement (repo README)
Installs on skills.sh: 12,043 (Nov 2, 2026)
GitHub stars: 11,140 (repo, Nov 2, 2026)
Files: 4 including SKILL.md: 2 prompt guides and Codex metadata, no scripts
Works in: Claude Code, Claude Agent SDK, Cursor, Windsurf, OpenAI-based agents and Codex, LangChain (README)
Last commit: Oct 28, 2026 (repo)

What it includes

Decides between T2VA, I2VA, FL2VA, L2VA and Ref2VA from your request.
Step 1 of the skill's workflow.
Pick the input mode
Builds the full audiovisual timeline from text alone.
Starts directly with the three core fields.
T2VA (text to video and audio)
Starts from the first frame and develops forward.
Opens with an instruction that says where the first picture is referenced in the target video.
I2VA (first frame)
Describes the continuous path between the two frames.
Aligns each picture to a time mark in the target video.
FL2VA (first and last frame)
Infers a plausible opening and converges to the supplied last frame.
The path lands on the image at the end of the clip.
L2VA (last frame)
Rewrites image, video and audio references into six sections.
Sections are subject_definitions, summary, retention_analysis, detailed_description, overall_soundscape and non_diegetic_music.
Ref2VA (full reference)
Guide for the four base modes, with shots, camera motion, dialogue and 4 cases.
Covers the prompt structure, keyframes, shots and cuts, camera motion, speakers and singing, on-screen text, soundscape and music.
references/base-en.txt
Guide for reference mode with label rules, retention analysis and a full example.
Defines <Subject N>, <Picture N>, <Video N> and <Audio N> labels.
references/ref-en.txt
Optional display metadata for the ChatGPT and Codex skills UI.
The frontmatter says it does not restrict the skill to OpenAI agents.
agents/openai.yaml

From the official docs

Questions

What does h3-prompt-writing do?

It makes your assistant write MiniMax H3 video prompts in H3's required structure. It supports text, first-frame, last-frame, first-and-last-frame and full-reference inputs.

How do I install h3-prompt-writing?

Use the install box above. The MiniMax-H3 README gives the skills CLI command npx skills add https://github.com/MiniMax-AI/MiniMax-H3 --skill h3-prompt-writing.

Does it generate the video?

No. It writes the prompt. You still need to run it in a MiniMax H3 tool. The skill's frontmatter says it needs no external API calls or MiniMax Hub tools.

Does it work outside Claude Code?

Yes, per the README: it lists Claude Code, the Claude Agent SDK, Cursor, Windsurf, OpenAI-based agents and Codex, LangChain, or any harness that can read a SKILL.md and local files.

What video length does it assume?

SKILL.md tips say to match the description's total duration to the requested video length, which it gives as 4 to 15 seconds.

Is it safe to install?

The folder holds SKILL.md, two text guides and one YAML metadata file, and bundles no scripts. The README says it makes no external API calls. Check the H3 license for your own use of the generated videos.

What about the other eight skills in the repo?

The README says the other eight are style-specific video generation skills built for the MiniMax Hub canvas workflow and are not portable to generic agent tools. This skill is the portable one.

Information

updated Nov 2, 2026

Repository: minimax-ai/minimax-h3
Folder: skills/h3-prompt-writing

minimax-ai/minimax-h3 skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 6

h3-prompt-writing SKILL.md · github.com
MiniMax-H3 README · github.com
h3-prompt-writing on skills.sh · skills.sh
MiniMax-AI/MiniMax-H3 · github.com
base-en.txt · github.com
ref-en.txt · github.com