In the development practice of DeepSeek Harness (DSH), enabling a model with visual, auditory, and video creation capabilities usually requires writing lengthy Skill prompts or manually maintaining complex scripts. The dsh-omni-workstation plugin turns this process into a tool, providing image analysis, generation, video synthesis, and speech processing through a unified configuration entry, and supports Any-to-Any full-modal interaction.

What Is This

dsh-omni-workstation is a full-modal workstation plugin that allows models to process video, image, and audio input and output, and supports calling ComfyUI image generation tools. The plugin is maintained by huashenglian, uses the MIT License, and belongs to the model inference plugin category.

Core Features

The plugin provides the following core capabilities:

  • VLM image analysis (analyze_image): Supports ordered multi-card failover and has built-in support for 28 providers.
  • Visual toolkit: Includes 6 tools, 4 of which are pure local tools with zero token consumption.
  • Image generation (generate_image): Supports OpenAI, DashScope, and ComfyUI protocols.
  • Multi-card asynchronous video generation (generate_video): Supports 7 protocols and provides an AI tool builder.
  • Speech synthesis and cloning (speak, clone_voice): Supports 3 cloud service providers and 4 local service providers.

Installation and Activation

Install the plugin using CLI commands:

dsh plugin --profile web add ./dsh-omni-workstation
dsh plugin --profile web add github:huashenglian/dsh-omni-workstation
dsh plugin --profile web add ./dsh-omni-workstation-0.1.0.tgz

After installation is complete, restart dsh web for the changes to take effect.

Typical Usage

  • Image analysis: Send an image (or a local path) to the AI and ask about its content. The system will automatically activate the analyze_image tool.
  • Video tool building: Enter /build-video-tool in the chat, and the AI will automatically build a custom video generation tool based on the instructions.
  • Configuration: Go to Settings -> Omni Workstation to configure it.

Notes

  • Avoid manual insertion: Do not manually add insertion lines in the configuration file, otherwise it will cause the duplicate loader entry id: omni-workstation error.
  • Card count limit: The default card count limit for video generation is 10.
  • Configuration storage: The configuration is stored in the omni-vision.json file inside the plugin installation directory. This file contains API keys, so do not commit it to version control.
  • Module switches: All modules can be independently enabled or disabled. After a module is disabled, its corresponding tool registration is completely removed, achieving 0 token cost.

Summary

By encapsulating multimodal capabilities into a plugin, dsh-omni-workstation reduces the maintenance cost for developers writing Skills or scripts. Its modular design allows features to be flexibly enabled or disabled according to requirements, making it suitable for scenarios that need to integrate visual, audio, and video processing capabilities in DeepSeek Harness.