dsh-comfyui-canvas

Building a ComfyUI workflow usually means two separate windows: one where the idea takes shape in chat, and another where you hunt for nodes, drag links, tweak parameters, and hit Run. The back-and-forth breaks your flow, and the AI agent can only emit a block of JSON for you to interpret manually instead of actually seeing what is happening on the canvas in front of you.

dsh-comfyui-canvas embeds the ComfyUI frontend as a split-screen canvas inside DeepSeek Harness, letting the agent operate the exact interface you are looking at: add nodes, wire links, set parameters, inject text, run workflows, and retrieve outputs, with every step rendered live on screen. It also covers batch parameter sweeps, automatic output-image retrieval back into the chat, one-click upgrades of the ComfyUI core and every custom node, and reading or editing custom-node source directly from the conversation.

This plugin is for creators who want an idea-to-finished-output pipeline across multi-modal generation, spanning images, music, video, and 3D. If you build and tune ComfyUI workflows and want an AI copilot that can see, edit, and drive the live canvas alongside your conversation, this removes the window-switching tax entirely.