dsh-omni-workstation

When working with multi-modal AI, the most common pain is scattered configuration: image analysis relies on a long Skill text block re-pasted into context every turn, video generation means hand-editing script paths and parameters, and swapping an API card requires another full re-paste. dsh-omni-workstation consolidates all of this into a single plugin — tools register into the Harness global registry, configuration lives in one JSON file, and changes take effect instantly without re-feeding instructions every session.

Core capabilities span five tracks: image analysis (ordered multi-card failover, 28 built-in providers, dynamic multimodal adaptation), six pure-local vision tools (zero-token zoom, color sampling, OCR, element detection, and more), image generation (OpenAI / DashScope / ComfyUI with multi-workflow management), multi-card async video generation (up to 10 cards, 7 protocols, plus /build-video-tool to let the AI assemble a custom tool on the fly), and voice synthesis with cloning (cloud MiMo / MiniMax / Doubao plus four local engines including IndexTTS and GPT-SoVITS). Each module has its own switch — turning one off completely unregisters its tools at zero token cost while preserving your config.

It is built for product managers, AI engineers, and independent creators working inside DeepSeek Harness who want a unified multi-modal workflow: fill in your cards and models in the settings page, and the AI immediately gains the ability to see images, generate art, produce video, and speak — without the mechanical chore of re-pasting recipes or hand-editing scripts.