hermes-plugin-minimax-video

Video generation plugin for Hermes Agent using MiniMax Hailuo models.

Brings MiniMax's Hailuo video models to Hermes's video_generate tool. Supports text-to-video, image-to-video, and subject-reference-to-video generation.

Models

Model	Modes	Strengths
MiniMax-Hailuo-2.3	T2V, I2V	Latest model. 768P/1080P, 6 or 10s. Camera control via [command] syntax.
MiniMax-Hailuo-2.3-Fast	I2V	Faster I2V variant. 768P/1080P, 6 or 10s.
MiniMax-Hailuo-02	T2V, I2V	Previous gen. 512P/768P/1080P, 6 or 10s. Camera control.
S2V-01	S2V	Subject-reference-to-video: animate a character from reference images.

Older models (T2V-01, T2V-01-Director, I2V-01, I2V-01-Director, I2V-01-live) are also supported -- pass them as the model parameter.

Install

hermes plugins install rriggs/hermes-plugin-minimax-video
hermes plugins enable minimax-video

Configure

All configuration lives under video_gen.minimax in config.yaml:

video_gen:
  provider: minimax
  minimax:
    api: https://api.minimax.io
    key_env: MINIMAX_PAYG_KEY
    model: MiniMax-Hailuo-2.3    # default for t2v/i2v
    model_s2v: S2V-01            # auto-used for subject reference
    # resolution: 768P           # optional
    # prompt_optimizer: true     # optional, defaults to true

Key	Default	Description
api	https://api.minimax.io	MiniMax API base URL
key_env	MINIMAX_API_KEY	Name of the env var holding your API key
model	MiniMax-Hailuo-2.3	Default model for T2V and I2V
model_s2v	S2V-01	Model for subject-reference-to-video
resolution	768P (Hailuo-2.3/02), 720P (others)	Video resolution
prompt_optimizer	true	Auto-optimize prompts

Set the API key in your .env file:

MINIMAX_PAYG_KEY=***

No env var names are hard-coded -- key_env tells the plugin which env var to read.

Usage

Once configured, Hermes's video_generate tool routes through MiniMax automatically:

• Text-to-video: just provide a prompt
• Image-to-video: provide image_url (used as the first frame)
  • Accepts URLs (https://...), data URIs (data:image/png;base64,...), or local file paths (/path/to/image.png) -- local files are automatically base64-encoded inline
• Subject-reference-to-video: provide reference_image_urls (the character to animate)

Camera Control

Hailuo-2.3, Hailuo-02, and Director models support camera movement commands embedded in the prompt using [command] syntax:

A man picks up a book [Pedestal up], then reads [Static shot].

Supported commands: [Truck left], [Truck right], [Pan left], [Pan right], [Push in], [Pull out], [Pedestal up], [Pedestal down], [Tilt up], [Tilt down], [Zoom in], [Zoom out], [Shake], [Tracking shot], [Static shot].

Multiple commands inside one [] take effect simultaneously (e.g. [Pan left,Pedestal up]). Sequential movements use then (e.g. ...[Push in], then...[Push out]).

Duration and Resolution

• Duration: 6 or 10 seconds (MiniMax only supports these values)
• Resolution depends on model:
  • Hailuo-2.3 / Hailuo-2.3-Fast: 768P (default), 1080P
  • Hailuo-02: 512P, 768P (default), 1080P
  • Other models: 720P (default)

Generated videos are cached locally under $HERMES_HOME/cache/videos/.

API Details

This plugin uses the MiniMax video generation API:

POST {api}/v1/video_generation          # Submit task
GET  {api}/v1/query/video_generation    # Poll status (?task_id=X)
GET  {api}/v1/files/retrieve            # Get download URL (?file_id=X)

All video generation is asynchronous. The plugin handles the full flow: submit, poll (5s interval, 10min timeout), retrieve file URL, download and cache locally.

Requirements

• Hermes Agent v0.18+
• API key set in the env var named by key_env
• requests Python package (bundled with Hermes)

License

MIT