dsh-plugin-use-images

The DeepSeek Harness runtime is built around a unified text-stream event model, whereas most image-generation APIs in the wild return a one-shot JSON payload or a raw binary stream. dsh-plugin-use-images bridges that protocol gap: pick an image model from the chat model selector, type a prompt, and the plugin forwards the request to your backend generation endpoint, converts the result into a native DSH image block, and persists it in the session attachment store so you can reference it in subsequent turns without ever leaving the conversation window.

Response handling is deliberately broad. The plugin recognises JSON URL arrays (data[].url, output.images[]), base64 fields (data[].b64_json, output.images[].b64_json), data: URIs, plain-text links, and direct binary responses carrying image/png, image/jpeg, image/webp, or image/gif. On the request side, the JSON body is a configurable template with {{prompt}}, {{model}}, {{n}}, and {{apiKey}} placeholders, and every header and extraction path is adjustable from the Web settings page. Changes hot-update the model list in place—no YAML editing, no restart.

This is the plugin to reach for when you run DSH as your daily AI chat interface and also have one or more image-generation backends (Stable Diffusion, Flux, DALL-E, or a custom pipeline) that you want tucked into the same conversation flow. It ships under the MIT licence, is implemented as a standalone LlmAdapter with zero modifications to DSH's text pipeline, and becomes active as soon as you restart dsh web after installing.