Generate Images Using Creative Director Pipeline

Generates images through a structured Creative Director pipeline instead of passing raw text to the API. Claude interprets what you need, selects the right domain expertise (Cinema, Product, Portrait, Editorial, UI, Logo, Landscape, Infographic, Abstract), and builds a detailed prompt using Google's 5-component formula so results reflect professional design conventions.

You run /banana generate with your idea. Claude asks clarifying questions if needed (use case, style, brand context), then checks for brand presets. It selects the domain mode that fits (for example, Product for packshots, Cinema for dramatic scenes), constructs a reasoning brief with Subject, Action, Location/Context, Composition, and Style components, chooses aspect ratio and resolution, and calls gemini-3.1-flash-image-preview. You get the image file path, the prompt Claude crafted, settings used, and suggestions for refinement.

Uses gemini-3.1-flash-image-preview by default. Claude reads prompt-engineering.md and gemini-models.md before every generation to ensure compliance with Google's API capabilities, banned keywords (8K, masterpiece, ultra-realistic), and prestigious context anchors. Errors like IMAGE_SAFETY filter blocks trigger up to 3 rephrase attempts with user approval. Rate limited to approximately 5-15 requests per minute (free tier). Python fallback script generate.py uses Google REST API directly if MCP unavailable.