dsh-minimax-image

English | 中文

A text-to-image plugin for DeepSeek Harness (DSH): turns a one-sentence visual description into a real image via the MiniMax image_generation API, saves it to your workspace, and exposes it to the model as an image-gen tool.

Features

• 🎨 Text-to-image: powered by the MiniMax image-01 model, with 8 aspect ratios (16:9 / 9:16 / 1:1 / 4:3 / 3:4 / 3:2 / 2:3 / 21:9)
• 🧩 DSH tool integration: registers the image-gen tool so the model can generate images directly from the conversation
• 🔑 Credential reuse: reads MINIMAX_API_KEY automatically (from the environment or the DSH credential store ~/.dsh/.credentials.yaml) — no extra setup
• 💾 Saved to disk: generated PNGs land in the workspace generated/ directory (customizable via the output_dir argument)

Installation

Local install:

dsh plugin --profile web add /path/to/dsh-minimax-image
dsh plugin --profile headless add /path/to/dsh-minimax-image

Usage

After restarting dsh web, just describe the picture in the conversation, for example:

Generate a cyberpunk city night scene with neon lights and rain

The model calls the image-gen tool and returns the saved file path:

Generated 1 image (16:9):
- D:\AI_video\generated\minimax-2026-08-18T12-16-11-1.png

Configuration

Config precedence (highest wins):

1. profile patch / plugin row config
2. ~/.config/dsh-minimax-image/config.json
3. MINIMAX_IMAGE_* environment variables
4. built-in defaults

Common options

Example ~/.config/dsh-minimax-image/config.json:

{
  "model": "image-01",
  "aspectRatio": "9:16",
  "outputDir": "generated",
  "apiKeyEnv": "MINIMAX_API_KEY"
}

Environment variables

License

MIT