Agent skill

Minimax H3 Creative Director

by unknowlei in unknowlei/minimax-h3-opencode-skills

Primary mandatory entrypoint for every MiniMax H3 video-generation request.

MITAuto-check passedMedia & Creative

Install Minimax H3 Creative Director

skills CLI
$ npx skills add unknowlei/minimax-h3-opencode-skills --skill minimax-h3-creative-director -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install unknowlei/minimax-h3-opencode-skills minimax-h3-creative-director --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/unknowlei/minimax-h3-opencode-skills.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/minimax-h3-creative-director .claude/skills/minimax-h3-creative-director && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
minimax-h3-creative-director
GitHub stars
122
Token cost
~2.8k tokens
SKILL.md length
1,461 words
Files
3 (incl. references)
Skills in repo
6
Repo updated
First seen
Licence
MIT

At a glance

Primary mandatory entrypoint for every MiniMax H3 video-generation request.

  • Works in 4 steps: Read ../h3-prompt-writing/SKILL.md first. → Inventory the request and supplied assets. → Read… → …
  • Tasks that involve AI video generation
  • SKILL.md covers Mandatory Official Read Order, Routing Decision, Direction Check and Multishot Gate, plus 3 more sections
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Minimax H3 Creative Director is an agent skill from unknowlei/minimax-h3-opencode-skills. Primary mandatory entrypoint for every MiniMax H3 video-generation request. Use before any other H3 skill for creation, animation, extension, editing, restyling, reference, or prompt review. It reads the official h3-prompt-writing specification, defaults image-based work to full-reference consistency, permits pure keyframe mode only for explicitly declared boundary-only images, and enforces a non-skippable multishot question workflow for explicit multishot requests or accepted 10-second-plus proposals before…

Its SKILL.md is about 2.8k tokens, which your agent loads only when the skill is triggered. The skill folder holds 4 other files, including reference files (for example `agents/openai.yaml` and `references/official-capabilities-and-routing.md`).

It sits in Media & Creative, covering AI video generation and Proposals and quotes. It works with MiniMax. The repository describes itself as: OpenCode skill suite for MiniMax H3 directing, routing, multishot planning, prompt generation, and review. The licence is MIT.

When your agent uses it

  • Tasks that involve AI video generation
  • Tasks that involve Proposals and quotes

Example prompts

  • “/minimax-h3-creative-director”

Workflow steps

4 steps, taken from the first numbered list in SKILL.md.

  1. Read ../h3-prompt-writing/SKILL.md first.
  2. Inventory the request and supplied assets.
  3. Read ../h3-prompt-writing/references/base-en.txt for T2VA, I2VA, FL2VA, or L2VA; read ../h3-prompt-writing/references/ref-en.txt for…
  4. Read references/official-capabilities-and-routing.md for official limits, asset roles, contradiction checks, and routing policy.

What it can do on your machine

Read from SKILL.md and the folder at commit 2e2096f. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Minimax H3 Creative Director loads about 2.8k tokens when it runs, and up to ~4.5k if it reads all its reference files. Until then it costs about 147 tokens; SKILL.md has 1,461 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~147
When it runs · the whole SKILL.md, loaded when a task matches
~2.8k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~4.5k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from unknowlei/minimax-h3-opencode-skills at commit 2e2096f, republished under its MIT licence (© unknowlei). 1,461 words, ~2,838 tokens.

Download SKILL.mdSave it as .claude/skills/minimax-h3-creative-director/SKILL.md (or your agent's skills folder). This skill also uses 2 other files; get the full folder from GitHub.
name
minimax-h3-creative-director
description
Primary mandatory entrypoint for every MiniMax H3 video-generation request. Use before any other H3 skill for creation, animation, extension, editing, restyling, reference, or prompt review. It reads the official h3-prompt-writing specification, defaults image-based work to full-reference consistency, permits pure keyframe mode only for explicitly declared boundary-only images, and enforces a non-skippable multishot question workflow for explicit multishot requests or accepted 10-second-plus proposals before routing to the final formatting specialist.

MiniMax H3 Creative Director

Act as the mandatory front door for MiniMax H3 work. Establish official format authority, choose the correct generation path, and invoke the matching specialist without making the user understand internal routing.

Mandatory Official Read Order

Before analyzing or drafting any H3 prompt:

  1. Read ../h3-prompt-writing/SKILL.md first.
  2. Inventory the request and supplied assets.
  3. Read ../h3-prompt-writing/references/base-en.txt for T2VA, I2VA, FL2VA, or L2VA; read ../h3-prompt-writing/references/ref-en.txt for reference generation, editing, continuation, or audio/reference work. Read both when routing is ambiguous or mixed.
  4. Read references/official-capabilities-and-routing.md for official limits, asset roles, contradiction checks, and routing policy.

Treat the official h3-prompt-writing files as canonical for prompt syntax. Treat the official-manual reference as canonical for product capabilities and operating limits. If a downstream skill conflicts with either authority, follow the official source.

Routing Decision

Choose exactly one final prompt-format specialist:

  • No reference assets, or text/script/storyboard only -> minimax-h3-text-video-prompt.
  • Pure keyframe exception -> minimax-h3-keyframe-video-prompt only when all conditions are true: the user explicitly calls each relevant image a literal first frame, last frame, or first-and-last-frame pair; the images serve only as boundary frames; and no image is expected to preserve or reference a character, identity, person, object, costume, scene, style, voice, action, camera, or other reusable trait.
  • Default for every other request containing an image -> minimax-h3-reference-video-prompt, including ambiguous image roles, ordinary image animation, character/person/object consistency, and any request where a boundary image also supplies reusable visual identity or content.
  • Mixed images/videos/audio, source-video editing, continuation, voice/audio control, or any other reference relationship -> minimax-h3-reference-video-prompt.
  • Existing prompt that must be checked, repaired, converted, or polished -> minimax-h3-prompt-reviewer.

minimax-h3-multishot-planner is an optional planning subskill, not a competing final specialist. Invoke it before the selected text, keyframe, or reference specialist when the multishot gate below requires it. Pass its confirmed shot plan to the final specialist as binding creative input.

Apply reference-consistency priority. Do not route to pure keyframe mode merely because there are one or two images, because the user says "animate this image," or because a first/last-frame interpretation seems plausible. If any pure-keyframe condition is absent or uncertain, route to full-reference without asking the user to downgrade. Full-reference can express a referenced image as a prompt-level first, key, or last frame while preserving character and object consistency.

After resolving creative questions and the multishot gate, read the selected final specialist's SKILL.md and required references, then execute it in the same task flow. Do not stop after announcing the route unless the user requested analysis only.

Direction Check

Use OpenCode's question tool before drafting whenever any condition below applies:

  • The request is short or supplies only a subject plus a basic action.
  • Two or more important creative axes are missing: exact duration, scene/event progression, visual medium/style, camera/edit rhythm, dialogue/voice, sound/music, or ending state.
  • A reference task does not clearly state desired action, scene progression, preservation priorities, audio treatment, or endpoint.
  • The user asks the AI to improvise, decide freely, surprise them, fill in the details, or otherwise choose the creative direction.
  • The user supplies an image with only a vague instruction such as "make it move" or "generate a video."
  • Whether source video should be edited, continued, or merely imitated
  • Whether audio should be reused exactly, partially reused, used as timbre/style reference, or replaced
  • Whether the user wants one continuous take or intentional cuts

Ask as many related questions as the current decision stage requires; normally use 1-5 in a general direction batch. Three is not a per-call, per-turn, or per-session ceiling. Recommend the strongest option and use the user's answers to complete the brief. Do not replace required questions with hidden assumptions. The only exception is an explicit instruction not to ask questions; then choose the most coherent interpretation and disclose material assumptions.

For every H3 choice question, a strict yes/no decision may use exactly two options. Every other choice question must provide at least five materially different, feasible options, with the strongest context-specific recommendation first and (Recommended) appended to its label. Explain the consequence of every option, keep each option label within five words, do not add Other, and never pad a list with near-duplicates. If five meaningful alternatives cannot be produced, ask an open-ended question instead of a short choice list.

Immediately before every question tool call, audit the proposed payload. If any non-binary question has fewer than five options, do not submit it; expand it to at least five meaningful choices or convert it to an open-ended question. This preflight is mandatory.

Do not ask whether an ambiguous image should use pure keyframe mode. The routing default already resolves that ambiguity in favor of full-reference consistency.

Multishot Gate

Resolve this gate after effective duration is known and before invoking the final prompt-format specialist:

  • If the user explicitly requests multiple shots, scenes, cuts, montage, a sequence, or shot-by-shot design, invoke minimax-h3-multishot-planner directly without asking whether multishot is desired.
  • If the user explicitly requests one continuous shot, one take, or no cuts, skip the multishot planner regardless of duration.
  • If duration is at least 10 seconds and the user has not expressed a single-shot or multishot preference, ask one binary question: whether to use a multishot structure. Recommend yes or no according to action density, continuity risk, and the user's goal rather than always recommending multishot.
  • If the user accepts, invoke minimax-h3-multishot-planner; if the user declines, record a single-shot constraint and continue.
  • If duration is below 10 seconds and no multishot preference is stated, do not ask this gate question and continue without the planner.
  • If duration is unknown, resolve duration first, then evaluate the gate. Exactly 10 seconds counts as at least 10 seconds.

The planner must first ask shot count, then plan every shot interactively. Once the user confirms its structured shot plan, pass that plan unchanged to the selected text, keyframe, or reference specialist. Do not repeat questions already answered during multishot planning.

Show full SKILL.md (492 more words)Show less
Non-skippable multishot state machine

Loading a Skill with the skill tool only places its instructions in context; it does not execute that Skill. Never treat skill({ name: "minimax-h3-multishot-planner" }) as completion of multishot planning.

When multishot is selected, enforce these states in order:

  1. multishot_selected: load minimax-h3-multishot-planner.
  2. planner_loaded: the next interactive action must establish shot count. If count was not explicitly supplied, the next tool call must be question; do not load a final specialist.
  3. shot_count_confirmed: plan Shot 1 through the final shot sequentially. The planner must issue the required per-shot question calls and receive answers for each shot.
  4. shots_confirmed: build the complete multishot_plan and ask the final binary confirmation.
  5. multishot_plan_confirmed: only now may a text, keyframe, reference, or reviewer Skill be loaded.

Hard blocker: while multishot is selected, do not load or invoke minimax-h3-text-video-prompt, minimax-h3-keyframe-video-prompt, minimax-h3-reference-video-prompt, or minimax-h3-prompt-reviewer until a complete confirmed multishot_plan exists in the conversation. If the planner was loaded but no shot questions were asked, remain in planner_loaded; do not advance.

Creative Brief

Before drafting, internally establish:

  • Asset-role map: upload order, stable label, intended role, and traits that must remain unchanged
  • Core creative line: subject, location, event/action, genre/style, and camera/edit strategy
  • Timeline: duration, shot boundaries, shot size, composition, action, camera, dialogue, sound effects, and endpoint
  • Preservation ledger: identity, clothing, objects, scene, text, voice, audio, and untouched source-video content
  • Delivery target: official online product/API or local ComfyUI workflow, because runtime limits may differ

Do not expose this internal brief unless it helps the user.

Mandatory Gates

  • Keep the official online target within 4-15 seconds, 24 FPS, and 7000 prompt characters. If a known local workflow has a narrower supported range, obey the runtime-specific range instead.
  • Assign an explicit role to every important uploaded asset. Never write vague references such as "use all references."
  • In pure keyframe mode, state each explicitly declared boundary role. In full-reference mode, a referenced image may be described in the prompt as a first, key, or last frame while still preserving its character/object/reference responsibilities.
  • Fit spoken words to the available shot time. Name on-screen versus off-screen speakers, preserve exact dialogue/lyrics, and state when speech continues across a cut. Use J-cut or L-cut only when its audio relationship is explicit.
  • Specify exact visible text, logo wording, captions, interface labels, or slogans; never paraphrase content that must appear verbatim.
  • Prefer observable action, composition, light, camera mechanics, and audible events over metaphors or adjective stacks.
  • For a physical orbit, use a mechanically coherent lateral camera move plus counter-pan when appropriate instead of relying only on the word "orbit."
  • Reject contradictions: one-take plus multiple cuts, music requested plus non_diegetic_music: N/A, face consistency without a usable identity reference, or excessive events for the duration.
  • H3 generates native stereo audio. Do not treat sound as an optional afterthought.

Response Policy

Return the specialist's copy-ready English prompt and complete Chinese translation. Keep routing commentary brief. Do not produce multiple competing prompt schemas unless the user explicitly asks for alternatives.

© unknowlei, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 2 other files (references) in skills/minimax-h3-creative-director of unknowlei/minimax-h3-opencode-skills.

  • SKILL.md
  • agents/openai.yaml
  • references/official-capabilities-and-routing.md

Open the folder on GitHubat commit 2e2096f

Compare with similar skills

Minimax H3 Creative Director next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Minimax H3 Creative Director compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Minimax H3 Creative Director this skillunknowlei/minimax-h3-opencode-skills122—~2.8kAutomated safety check: PassMIT
Minimax H3 Video Reversegnipbao/minimax-h3-video-reverse-skill230—~2.2kAutomated safety check: PassMIT
Minimax H3 Colabkillkli/minimax-h3-colab-skill110—~743Automated safety check: PassNone
H3 Prompt Writingvllm-project/vllm-omni7.1k6 repos~744Automated safety check: PassApache-2.0
Handdrawn Live Video Generatortl2012tl/comfyUI-llama-TE2414 repos~1.5kAutomated safety check: PassNone
ComfyUI Local DriverSlavaSexton/ComfyUI-Agent-Kit105—~12kAutomated safety check: PassApache-2.0

Similar skills

  • Minimax H3 Video Reverse

    gnipbao/minimax-h3-video-reverse-skill

    先复述参考视频内容与叙事逻辑,和用户确认理解后,再反推 MiniMax H3 文生视频或逐镜图生视频提示词。Use when the user asks to reverse-engineer a video into prompts, reconstruct shots, extract first/end-frame prompts, adapt a reference to H3, or…

    230 GitHub stars~2.2k tokensUpdated 8 days ago
    Media & CreativeAuto-check passed
  • Minimax H3 Colab

    killkli/minimax-h3-colab-skill

    Create short MiniMax H3 reference-to-video clips from one or more local images through Google Colab CLI.

    110 GitHub stars~743 tokensUpdated 14 days ago
    Media & CreativeAuto-check passed
  • H3 Prompt Writing

    vllm-project/vllm-omni

    Write MiniMax H3 video generation prompts for T2VA, I2VA, FL2VA, L2VA, and Ref2VA.

    7.1k GitHub starsUsed in 6 repos~744 tokens
    Media & CreativeAuto-check passed
  • Handdrawn Live Video Generator

    tl2012tl/comfyUI-llama-TE

    For creators making surreal short videos that blend rough glowing hand-drawn animation with live-action spaces.

    241 GitHub starsUsed in 4 repos~1.5k tokens
    Media & CreativeAuto-check passed
  • ComfyUI Local Driver

    SlavaSexton/ComfyUI-Agent-Kit

    Drives a local ComfyUI install over its HTTP API to generate and edit images, video and audio, with per-model prompt recipes and workflow guidance.

    105 GitHub stars~12k tokensUpdated 1 mo ago
    Media & CreativeAuto-check passed
  • Mmx CLI

    agentscope-ai/OpenJudge

    Generate text, images, video, speech, and music via the MiniMax AI platform.

    871 GitHub stars~655 tokensUpdated 1 mo ago
    Media & CreativeAuto-check passed

More from unknowlei/minimax-h3-opencode-skills

  • Minimax H3 Multishot Planner

    unknowlei/minimax-h3-opencode-skills

    Non-skippable planning-only MiniMax H3 subskill used by minimax-h3-creative-director before final prompt formatting.

    122 GitHub stars~2.2k tokensUpdated 2 mo ago
    Auto-check passed
  • Minimax H3 Keyframe Video Prompt

    unknowlei/minimax-h3-opencode-skills

    Narrow downstream MiniMax H3 specialist for pure I2VA, FL2VA, and L2VA boundary-frame prompts.

    122 GitHub stars~2.4k tokensUpdated 2 mo ago
    Auto-check passed
  • Minimax H3 Reference Video Prompt

    unknowlei/minimax-h3-opencode-skills

    Default downstream MiniMax H3 specialist for every image-based request unless the user explicitly declares boundary-only first/last frames with no reusable reference role.

    122 GitHub stars~3k tokensUpdated 2 mo ago
    Auto-check passed
  • Minimax H3 Text Video Prompt

    unknowlei/minimax-h3-opencode-skills

    Downstream MiniMax H3 specialist for professional text-to-video (T2VA) prompts using the official three-field format.

    122 GitHub stars~2.3k tokensUpdated 2 mo ago
    Auto-check passed
  • Minimax H3 Prompt Reviewer

    unknowlei/minimax-h3-opencode-skills

    Downstream MiniMax H3 specialist that audits, repairs, and rewrites T2VA, I2VA, FL2VA, L2VA, and full-reference prompts into an official structured format.

    122 GitHub stars~2.1k tokensUpdated 2 mo ago
    Auto-check passed

Works with

Questions about Minimax H3 Creative Director

What does Minimax H3 Creative Director do?

Primary mandatory entrypoint for every MiniMax H3 video-generation request. Minimax H3 Creative Director is an agent skill from unknowlei/minimax-h3-opencode-skills. Primary mandatory entrypoint for every MiniMax H3 video-generation request.

When should I use Minimax H3 Creative Director?

Minimax H3 Creative Director fits situations like: tasks that involve AI video generation; tasks that involve Proposals and quotes.

How do I install Minimax H3 Creative Director in Claude Code?

Run `npx skills add unknowlei/minimax-h3-opencode-skills --skill minimax-h3-creative-director -a claude-code`. Or copy the skill folder (skills/minimax-h3-creative-director in unknowlei/minimax-h3-opencode-skills) into .claude/skills/minimax-h3-creative-director in your project. Claude Code loads it when a task matches its description.

How do I install Minimax H3 Creative Director in Codex?

Run `npx skills add unknowlei/minimax-h3-opencode-skills --skill minimax-h3-creative-director -a codex`. Or copy the skill folder (skills/minimax-h3-creative-director in unknowlei/minimax-h3-opencode-skills) into .agents/skills/minimax-h3-creative-director in your project. Codex loads it when a task matches its description.

Can I use Minimax H3 Creative Director in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add unknowlei/minimax-h3-opencode-skills --skill minimax-h3-creative-director -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/minimax-h3-creative-director, .gemini/skills/minimax-h3-creative-director, .github/skills/minimax-h3-creative-director and .opencode/skills/minimax-h3-creative-director in your project.

What does Minimax H3 Creative Director need to run?

SKILL.md names no scripts, command-line tools or credentials: Minimax H3 Creative Director is instructions for the agent only.

Does Minimax H3 Creative Director access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Minimax H3 Creative Director safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Minimax H3 Creative Director use?

Minimax H3 Creative Director is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Minimax H3 Creative Director use?

About 2.8k tokens (SKILL.md is roughly 11k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 1.6k tokens, read only when the agent opens those files.

What are the alternatives to Minimax H3 Creative Director?

Skills that share tags, products or a category with Minimax H3 Creative Director: Minimax H3 Video Reverse (gnipbao/minimax-h3-video-reverse-skill, 230 stars), Minimax H3 Colab (killkli/minimax-h3-colab-skill, 110 stars), H3 Prompt Writing (vllm-project/vllm-omni, 7.1k stars) and Handdrawn Live Video Generator (tl2012tl/comfyUI-llama-TE, 241 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Minimax H3 Creative Director?

unknowlei (a GitHub user) maintains it in unknowlei/minimax-h3-opencode-skills, which has 122 GitHub stars. The repository holds 6 skills in this directory. The repository was last updated on August 9, 2026.

Source: unknowlei/minimax-h3-opencode-skills on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.