dsh-image2-draw

DSH has no built-in image generation entry point, so calling gpt-image-2 means leaving the chat and hand-crafting API requests. dsh-image2-draw plugs text-to-image and image-to-image generation directly into the conversation: configure any OpenAI Images-compatible relay with just a base URL and an API key, and the resulting pictures appear as an in-chat tool card with thumbnails, zoom, and Save As, while also landing in outputs/image2/.

Key capabilities include up to 8 relay endpoints with automatic failover on network errors, timeouts, and 401/402/403/404/429/5xx responses (parameter errors like 400 never trigger a switch); text-to-image producing 1 to 8 images per call and image-to-image accepting 1 to 8 PNG/JPEG/WebP references; and a docked studio above the input box with tabs for text-to-image, image-to-image, and multi-view character mode, where uploading 2 to 8 photos of the same person from different angles yields a new shot that preserves face, build, and hairstyle. Built-in character-design Sheet A/B templates and a persistent preset library with variable filling round out the workflow.

This plugin suits developers and creators who already use DeepSeek Harness for daily work and want to generate concept art, character sheets, or outfit mock-ups without leaving the chat. The same repository also ships as a Claude Code skill sharing the identical core implementation, so both environments work seamlessly.