If you need text rendered accurately in generated images,product labels, multilingual signage, UI mockups with real headlines,this is the model that actually does it. Routes to OpenAI's GPT Image 2 via RunComfy's API, supports three fixed aspect ratios, and handles multi-reference edits that preserve composition while you tweak one thing at a time. The prompting guide is refreshingly honest about what works: quote your embedded text exactly, use compositional cues like "rule of thirds" directly, don't pile up conflicting styles. Best for brand-safe commercial work where layout precision beats artistic flourish. If you want painterly or cinematic instead, the skill steers you to Flux or Seedream without forcing the issue.
npx skills add https://github.com/agentspace-so/runcomfy-agent-skills --skill gpt-image-2