Generate Images Using Creative Director Pipeline2 skills
Claude orchestrates intent analysis, domain selection, and 5-component prompt construction to create optimized AI images.
Content Blueprint
Safety checkedGenerate AI Images Using Claude as Creative Director
AgriciDanielMay 2026
A Claude Code plugin that orchestrates Google's Gemini models for AI image generation, handling intent analysis, domain selection, and prompt engineering.
Creative DirectionImage GenerationPrompt Engineering
Install Command
Run this command to deploy the blueprint to your environment.
What problem does this solve?
Turns Claude into a Creative Director that interprets your intent and constructs optimized prompts instead of passing raw text to image generators. This avoids generic results and ensures each image generation pulls from domain expertise, proper prompt structure, and design best practices.
How does it work?
Once installed, you run /banana generate with your idea. Claude analyzes what you need (blog header, product shot, portrait), selects a domain mode (Cinema, Product, Portrait, Editorial, UI, Logo, Landscape, Infographic, Abstract), constructs a detailed prompt using Google's 5-component formula (Subject, Action, Location/Context, Composition, Style), sets aspect ratio and resolution, then calls Gemini to generate. Output: optimized images with the reasoning shown so you understand what was generated and why.
Included Skill Groups
What's the biggest win?
Every image generation respects domain conventions and composition rules instead of relying on generic AI defaults, so you get professional results on the first try.
What should I know technically?
Requires Node.js 18+ for npx plugin installation and a free Google AI API key from https://aistudio.google.com/apikey. Default model is gemini-3.1-flash-image-preview with up to 4K output resolution (512, 1K, 2K, 4K configurable). The skill uses Python 3 fallback scripts (generate.py, edit.py) to call Gemini REST API directly if MCP is unavailable, so no additional runtime dependencies for basic operation. Cost tracking and brand presets store data locally in ~/.banana/. ImageMagick (magick or convert command) is optional and only needed if you use post-processing features (transparent backgrounds, format conversion, resizing).
What should I watch out for?
Free tier has reduced rate limits (approximately 5-15 requests per minute, approximately 20-500 per day) as of late 2025. Google's safety filters may block prompts containing violence, explicit content, or real public figures. The skill offers rephrase strategies to work around false positives. ImageMagick is optional but required for post-processing (background removal, format conversion, cropping). Batch API (50% cost discount) requires non-urgent bulk workflows. Some advanced features like search grounding and multi-image references depend on MCP package version support.
Key Features
9 Domain Modes
Cinema, Product, Portrait, Editorial, UI/Web, Logo, Landscape, Infographic, Abstract. Each has specialized prompt construction rules and visual conventions.
5-Component Prompt Formula
Google's official formula (Subject, Action, Location/Context, Composition, Style) ensures detailed, structured prompts that yield consistent results.
Intelligent Image Editing
/banana edit enhances user instructions into detailed editing briefs (edge preservation, color accuracy, placement precision) before calling the edit API.
Batch Variations
/banana batch generates N variations by rotating components (lighting, composition, style) so you compare different approaches in one workflow.
Multi-Turn Creative Sessions
/banana chat maintains character consistency and style across refinement turns without losing context between edits.
Tools in this Blueprint
About This Blueprint
- License
- MIT
- Industry
- Technology
- Skills
- 1 skills total