GPT Image 2 Enhanced

This skill documents all enhancements made to the base gpt-image-2 skill. It provides:

• Dual API support (OpenAI + Gemini style)
• 3 new custom templates
• Complete template catalog (17 categories / 99 templates)
• Scenario-based quick lookup
• Correct API calling format for Gemini-style endpoints

API Support

Gemini-Style API (e.g., nn.147ai.com)

The enhanced generate.js auto-detects Gemini-style APIs via isGeminiApi() in shared.js. Detection triggers when OPENAI_BASE_URL contains nn.147ai.com or API_STYLE=gemini env var is set.

Endpoint:

POST https://nn.147ai.com/v1beta/models/gpt-image-2:generateContent

Request format:

{
  "contents": [{
    "parts": [{"text": "your image prompt here"}]
  }],
  "config": {
    "responseModalities": ["TEXT", "IMAGE"],
    "imageConfig": {
      "aspectRatio": "16:9",
      "imageSize": "2K"
    }
  }
}

Headers:

Authorization: Bearer YOUR_API_KEY
Content-Type: application/json

Response format: Image embedded as data:image/png;base64,... text in JSON. Extracted via regex data:image/(\w+);base64,([A-Za-z0-9+/=]+).

Aspect ratios: 1:1, 9:16, 16:9, 3:4, 4:3, 3:2, 2:3, 5:4, 4:5, 21:9

Image sizes: 1K, 2K, 4K

OpenAI-Style API

Standard /v1/images/generations endpoint. Used when OPENAI_BASE_URL is a standard OpenAI-compatible endpoint.

Modified Scripts

shared.js additions:

• isGeminiApi() — detects if current API is Gemini-style
• buildGeminiPayload(prompt, cfg) — builds Gemini-format request body
• buildGeminiUrl() — builds Gemini endpoint URL
• postGeminiJson(url, payload) — POSTs to Gemini endpoint
• extractGeneratedBytes(json) — updated to parse both OpenAI and Gemini responses

generate.js changes:

• Auto-routes between Gemini and OpenAI API styles
• --size maps to Gemini aspectRatio values
• --quality maps to Gemini imageSize (high→2K, default→1K)

Environment variables:

OPENAI_API_KEY=your_key
OPENAI_BASE_URL=https://nn.147ai.com/v1
ENABLE_GARDEN_IMAGEGEN=1

New Templates

1. knowledge-card (知识卡片)

Path: references/infographics/knowledge-card.md

Purpose: Upload a document/article, auto-extract key points into visual knowledge cards.

Variants:

• Standard: warm white bg + blue/orange accents, 4 key points
• Tech docs: dark background + cyan/green, code snippets
• WeChat style: Morandi palette, literary, WeChat-optimized

2. article-illustration (新媒体配图)

Path: references/slides-and-visual-docs/article-illustration.md

Purpose: Generate social media illustrations from text content.

Variants:

• WeChat cover: 16:9, bold title, simple visual
• Xiaohongshu: 3:4, lifestyle aesthetic, warm tones
• Tech blog: dark bg, code/green accents

3. style-transfer-poster (风格迁移海报)

Path: references/poster-and-campaigns/style-transfer-poster.md

Purpose: Reference one image's visual style to create new theme posters.

Variants:

• Brand extension: extend main visual across materials
• Style copy + theme replace: 95%+ style fidelity
• Cross-medium: apply paint style to UI, or illustration to poster

Complete Template Catalog

01. Academic Figures (9 templates)

references/academic-figures/

02. Assets & Props (2)

references/assets-and-props/

03. Avatars & Profile (5)

references/avatars-and-profile/

04. Branding & Packaging (6)

references/branding-and-packaging/

05. Editing Workflows (5)

references/editing-workflows/

06. Grids & Collages (5)

references/grids-and-collages/

07. Infographics (7)

references/infographics/

08. Maps (5)

references/maps/

09. Portraits & Characters (5)

references/portraits-and-characters/

10. Poster & Campaigns (9)

references/poster-and-campaigns/

11. Product Visuals (6)

references/product-visuals/

12. Scenes & Illustrations (4)

references/scenes-and-illustrations/

13. Slides & Visual Docs (5)

references/slides-and-visual-docs/

14. Storyboards & Sequences (8)

references/storyboards-and-sequences/

15. Technical Diagrams (7)

references/technical-diagrams/

16. Typography & Text Layout (2)

references/typography-and-text-layout/

17. UI Mockups (6)

references/ui-mockups/

Scenario-Based Quick Lookup

Academia / Education

• Method pipeline → method-pipeline-overview
• Neural network diagram → neural-network-architecture
• Experimental setup → scientific-schematic
• Journal Graphical Abstract → graphical-abstract
• Thesis defense poster → research-overview-poster
• Textbook diagram → educational-diagram-slide

Branding / Marketing

• Brand identity system → brand-identity-board
• Campaign Key Visual → campaign-kv
• Web banner / hero → banner-hero
• Magazine cover → editorial-cover
• KPI dashboard → kpi-dashboard-infographic
• Style transfer → style-transfer-poster

E-commerce / Products

• Product exploded view → exploded-view-poster
• White background product → white-background-product
• Premium studio shot → premium-studio-product
• Lifestyle scene → lifestyle-product-scene
• E-commerce board → ecommerce-marketing-board
• Background replacement → background-replacement

Social Media / Content

• Article cover image → article-illustration
• Knowledge card → knowledge-card
• Hand-drawn infographic → hand-drawn-infographic
• Step-by-step tutorial → step-by-step-infographic
• Professional headshot → professional-portrait
• Sticker pack → sticker-set

Travel / Maps

• Food map → food-map
• Travel route → travel-route-map
• City illustration → illustrated-city-map
• Store distribution → store-distribution-map
• Day trip itinerary → itinerary-day-trip-map

Film / Animation / Comics

• 4-panel comic → four-panel-comic
• Manga spread → manga-spread-page
• Anime Key Visual → anime-key-visual
• TVC storyboard → product-tvc-storyboard
• Cinematic storyboard → cinematic-storyboard-grid
• Healing scene → healing-scene

Technology / Development

• System architecture → system-architecture
• Flowchart / decision → flowchart-decision
• Sequence diagram → sequence-diagram
• State machine → state-machine
• ER diagram → er-diagram
• Mind map → mind-map-tech
• Network topology → network-topology

Design / IP / Creative

• Mascot brand kit → mascot-brand-kit
• Character design sheet → character-sheet
• Pose reference → pose-reference-sheet
• Character grid portrait → character-grid-portrait
• Anime pitch board → anime-pitch-board
• Icon set → retro-skeuomorphic-icons

Image Editing

• Background swap → background-replacement
• Object replacement → local-object-replacement
• Remove clutter → object-removal
• Product retouching → product-retouching
• Portrait edit → portrait-local-edit

Batch Generation

To batch generate images, create a JSON array and use the Node.js script pattern:

import { readFile, writeFile, mkdir } from 'node:fs/promises';

const API = 'https://nn.147ai.com/v1beta/models/gpt-image-2:generateContent';
const KEY = process.env.OPENAI_API_KEY;

const entries = [
  { cat: "category", tmpl: "template-name", prompt: "your prompt" },
  // ...
];

for (const e of entries) {
  const body = {
    contents: [{ parts: [{ text: e.prompt }] }],
    config: { responseModalities: ["TEXT", "IMAGE"], imageConfig: { aspectRatio: "16:9", imageSize: "1K" } }
  };
  // ... fetch, extract base64, save
}

Quick Reference