Got it—here’s a tight, production-oriented “master prompt + plan” for your icon-generation subsystem, built around **Gemini (to write the per-word prompts)** and **ComfyUI (to render the images)**. I’ve split it into discrete tasks so each can be delivered independently.
