dsh-jimeng

DeepSeek Harness is a text-first conversation tool, which means generating images always meant leaving the chat. dsh-jimeng bridges that gap by plugging Volcengine Ark's Jimeng (Seedream) model directly into your conversation—type "draw a whale" and the AI calls the model, saves the image locally, registers it as a chat attachment, and renders it inline. No extra tabs, no manual file handling.

The plugin exposes two workflows. As an AI tool, it auto-invokes generate_image (text-to-image) or image_to_image (image-to-image; reference images accept local paths, URLs, or data URLs). As a settings panel under "Jimeng Image", you hand-write the prompt, pick a model from a built-in Seedream dropdown (or auto-detect available models), adjust size, toggle watermark, and hit generate. Because it speaks the OpenAI-compatible /images/generations API, it also works with SiliconFlow, custom relays, or any other endpoint—just swap the baseUrl and model name. API keys stay server-side; the browser only sees a masked value.

It is built for anyone who lives in DeepSeek Harness for daily creative work—concept sketches, product mockups, quick visual ideas—and for developers who want to consolidate multiple image-generation backends behind one conversational UI. All you need is a Volcengine Ark API key (or any compatible endpoint) and the plugin is ready to go.