메뉴
Generate images using Google's Nano Banana 2 (Gemini 3.1 Flash Image) with workflow-based prompting
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | image-gen |
| description | Generate images using Google's Nano Banana 2 (Gemini 3.1 Flash Image) with workflow-based prompting |
| triggers | ["create image","generate image","make infographic","create infographic","generate diagram","make diagram","design visual","create visual"] |
| allowed-tools | Read, Write, Bash |
| version | 0.1.0 |
Generate professional images, infographics, and diagrams using Google's Nano Banana 2 model (gemini-3.1-flash-image-preview).
Nano Banana 2 (released February 26, 2026):
All scripts use Python via uv run with inline dependencies.
uv run scripts/generate.py "prompt" output.png [aspect_ratio] [size]
Examples:
# Basic image
uv run scripts/generate.py "A cozy coffee shop in autumn" coffee.png
# Infographic with specific aspect ratio
uv run scripts/generate.py "Infographic explaining how neural networks work" nn.png 16:9 2K
# 4K professional image
uv run scripts/generate.py "Professional headshot, studio lighting" headshot.png 3:2 4K
uv run scripts/edit.py input.png "edit instructions" output.png
Examples:
# Edit existing image
uv run scripts/edit.py photo.png "Change the background to a beach sunset" edited.png
uv run scripts/compose.py "prompt" output.png --refs image1.png image2.png
Examples:
# Combine styles from multiple images
uv run scripts/compose.py "Combine these styles into a logo" logo.png --refs style1.png style2.png
Workflows provide structured approaches for specific visual types. Each workflow follows the PAI 6-step editorial process:
When generating images, follow the appropriate workflow:
1. What data/concept needs visualization?
2. What's the key insight or takeaway?
3. Aspect ratio: 16:9 (landscape) recommended
4. Include: clear hierarchy, minimal text, supporting icons
5. Generate at 2K minimum for text clarity
1. What system/process is being illustrated?
2. What are the key components and relationships?
3. Style: flat colors, clean lines, minimal detail
4. Generate at 2K for label clarity
Requires GEMINI_API_KEY environment variable. This should be set from Geoffrey's secrets:
source ~/Library/Mobile\ Documents/com~apple~CloudDocs/Geoffrey/secrets/.env
By default, save images to /tmp/ or user-specified paths. For persistent storage, use:
~/Library/Mobile Documents/com~apple~CloudDocs/Geoffrey/images/
DO NOT use the Read tool on generated images.
Why:
Workflow:
| Size | Cost per Image |
|---|---|
| 1K | Free tier / $0.04 |
| 2K | $0.134 |
| 4K | $0.24 |
Automate the version bump ritual — three independent tracks (marketplace, psd-coding-system, psd-productivity)
Auto-evolve the plugin — compounds learnings into CLAUDE.md/patterns/agents then prunes them, checks releases, compares competition, contributes patterns
Autonomous build-to-done — implement, verify the full Definition of Done (build/lint/typecheck/full suite/Playwright + screenshots), open a PR with visual evidence, then watch CI and the AI reviewers and fix every round until 100% clean. Does not stop until done.
Clarify intent, research in parallel, design the approach, and emit a task breakdown + a machine-checkable Definition of Done — optionally filing contract-compliant GitHub issues that /lfg can pick up and drive to done.
Sign, notarize, and package a macOS .app into a .pkg for PSD Jamf Self Service deployment — full Apple Developer ID pipeline (sign → notarize → staple → pkgbuild → notarize pkg → verify). Triggers on "sign app", "notarize app", "package for jamf", "build pkg".
Configure this project's verification gate and review agents — writes .psd/verify.json (build/lint/test/typecheck/e2e commands, E2E flows, strictness, AI-reviewer logins, learnings, active review agents)