Wraps Google's flash-tier Gemini image model with documented prompting patterns that actually work: subject-first grammar, quoted text for in-image typography, seed locking for reproducible variants. Calls RunComfy's CLI for batching drafts at 0.5K or pushing finals to 2K. The bundled guide knows when to route to Nano Banana Pro for portraits or Flux/GPT Image for stylization, which matters because the skill triggers on generic asks like "gemini image" or "google image gen." The web grounding option is there for current-event imagery but adds latency. Best for social thumbnails, rapid ideation rounds, and anything where you need predictable framing over maximum detail.
npx skills add https://github.com/agentspace-so/runcomfy-agent-skills --skill nano-banana-2