dsh-imagegen-plugin

DeepSeek Harness main models are built for reasoning, not for painting. When your workflow demands a quick illustration, a background swap, or a masked repaint, jumping to a browser tab or a CLI script and pasting the result back breaks the flow. dsh-imagegen-plugin closes that gap by turning image generation into a first-class, conversation-invokable skill inside DSH: the main model orchestrates the call, and the image appears inline in the same thread.

The plugin speaks four endpoint dialects out of the box: OpenAI-compatible /v1/images (covering OpenAI, OpenRouter, SiliconFlow, and similar providers), Alibaba DashScope async tasks, Replicate, and fal.ai. It supports both text-to-image and image-editing tasks such as masked inpainting and instruction-based repaints. Configuration is a one-time fill-in on a settings card (endpoint, model ID, API key, default size and count), or a single chat sentence that copies the base URL and key from the main model with one call. Generated files are saved to disk with an absolute path, rendered as Markdown in the conversation for instant viewing, and fed back into model context so the next turn can reference and iterate on the same image.

If you already run DSH for daily development or content work and hold API access to one or more image-generation providers, this plugin removes the copy-download-reupload loop: describe what you want, see the picture in the same thread, and keep refining without leaving the conversation.