Minimax Tools Skill

MiniMax Tools Skill

An OpenClaw skill for calling MiniMax multimodal APIs directly through local Python wrappers.

Instead of relying on the official MiniMax MCP server, this repository provides a maintained, script-first skill that is easy to use from OpenClaw, shell workflows, and local automation.

What this skill does

This skill currently supports:

• Speech synthesis (TTS)

  • synchronous TTS
  • built-in default Chinese and English voices
  • explicit voice_id support
  • speed / volume / pitch / emotion controls

• Voice cloning

  • upload clone audio
  • upload optional prompt audio
  • create cloned voices
  • optionally download preview audio (demo_audio)
  • one-shot clone-from-files workflow

• Image generation

  • text-to-image
  • reference-image generation via subject_reference
  • local image path / URL / Data URL support

• Video generation

  • text-to-video
  • image-to-video
  • first/last-frame video generation
  • async task query / wait / download flow

• Music generation

  • prompt + lyrics song generation
  • instrumental generation
  • lyrics optimizer support

What is not implemented yet

At the moment, this repository does not include:

• subject-reference-to-video (s2v)
• video agent templates
• voice design
• voice listing / management helpers
• async long-text TTS wrappers
• full MiniMax file management coverage beyond what this skill needs today

Why this skill exists

MiniMax exposes a broad multimodal API surface, but not every workflow is convenient to use through raw HTTP or an aging MCP integration.

This skill is designed for:

• direct local control
• predictable file outputs
• reproducible script-based workflows
• easy reuse inside OpenClaw skills and automations

Repository structure

.
├── SKILL.md
├── README.md
├── LICENSE
├── references/
│   └── api-notes.md
└── scripts/
    ├── common.py
    ├── minimax.py
    ├── minimax_tts.py
    ├── minimax_voice.py
    ├── minimax_image.py
    ├── minimax_video.py
    └── minimax_music.py

Requirements

Runtime

• Python 3
• requests

Install dependency if needed:

pip install requests

Environment variables

Required:

export MINIMAX_API_KEY=your_api_key_here

Optional:

export MINIMAX_BASE_URL=https://api.minimaxi.com

Install

From ClawHub

clawhub install minimax-tools

From GitHub

Clone this repository and copy or symlink it into your OpenClaw skills/ directory.

Example:

ln -s /home/yangtao/repos/minimax-tools-skill <workspace>/skills/minimax-tools

Unified CLI

Use the single entrypoint:

python3 scripts/minimax.py <subcommand> ...

Available top-level subcommands:

• tts
• voice
• image
• video
• music

Quick examples

TTS with default Chinese voice

python3 scripts/minimax.py tts \
  --text "你好，欢迎使用 MiniMax。"

TTS with default English voice

python3 scripts/minimax.py tts \
  --text "Hello, welcome to MiniMax." \
  --voice-lang en

Clone a voice in one step

python3 scripts/minimax.py voice clone-from-files \
  --clone-file ./voice.wav \
  --voice-id MyVoice001 \
  --text "你好，欢迎使用我的复刻音色。" \
  --download-demo

Text-to-image

python3 scripts/minimax.py image \
  --prompt "Cyberpunk city at night, cinematic lighting, dense details" \
  --aspect-ratio 16:9

Reference-image generation

python3 scripts/minimax.py image \
  --prompt "Same character standing in a rainy neon street" \
  --subject-reference ./character.png \
  --aspect-ratio 16:9

Text-to-video

python3 scripts/minimax.py video create \
  --prompt "An orange cat watching rain by the window, cinematic, soft light" \
  --wait --download

Image-to-video

python3 scripts/minimax.py video i2v \
  --first-frame-image ./cat.png \
  --prompt "Slow push-in, the cat blinks, rain outside the window" \
  --wait --download

First/last-frame video

python3 scripts/minimax.py video fl2v \
  --first-frame-image ./young.jpg \
  --last-frame-image ./old.jpg \
  --prompt "A person slowly grows older, stable camera" \
  --wait --download

Music generation

python3 scripts/minimax.py music \
  --prompt "indie folk, warm, late night city walk" \
  --lyrics "[Verse]\nNight wind slips across the street lights..."

TTS defaults

Current built-in defaults:

• Chinese voice: female-tianmei
• English voice: English_Graceful_Lady
• Model: speech-2.8-turbo
• Format: mp3
• Sample rate: 32000
• Bitrate: 128000

Voice cloning constraints

Clone source audio

• formats: mp3, m4a, wav
• duration: 10 seconds to 5 minutes
• size: <= 20 MB

Prompt audio

• formats: mp3, m4a, wav
• duration: under 8 seconds
• size: <= 20 MB

MiniMax docs note that cloned voices are temporary unless used in real TTS within 7 days.

Image generation notes

The current image wrapper supports:

• text-to-image
• reference-image generation through subject_reference

This is closer to reference-driven generation / character consistency than a full mask-based editor workflow.

Video generation notes

Video generation is asynchronous. The typical flow is:

1. create a task
2. query or wait for completion
3. download the final file

The wrapper already supports the full create/query/wait/download path.

Output behavior

Scripts print JSON to stdout for easy automation.

Typical fields include:

• ok
• trace_id
• voice_id
• task_id
• file_id
• id
• path
• paths
• meta

ClawHub publishing notes

This skill declares its required environment variable in SKILL.md metadata:

• MINIMAX_API_KEY

That helps ClawHub understand the skill’s external requirements and reduces the chance of publish-time flagging.

License

MIT-0