GPT-Image2 Architecture and Space Templates: Complete Guide with JSON and Plain-Text Formats

GPT-Image2 provides dual-format templates for architectural visualization that include both human-readable plain-text prompts and machine-parseable JSON schemas, enabling precise control over space type, materials, camera angles, and rendering quality.

The freestylefly/awesome-gpt-image-2 repository ships with a dedicated "建筑与空间" (Architecture & Spaces) template family designed specifically for generating photorealistic architectural renders. These templates for architecture and spaces in GPT-Image2 are defined in docs/templates.md and offer fine-grained control over environmental context, lighting scenarios, and material specifications while allowing the model to handle creative details like furniture layout.

Template Structure and Schema Overview

Located at docs/templates.md#tpl-architecture, the architecture template system provides two interchangeable formats that share an identical semantic schema. You can choose between a plain-text prompt for chat-based interactions or a JSON-structured prompt for programmatic API calls.

Plain-Text Template Format

The human-readable variant uses concise Chinese placeholders ideal for direct paste into chat interfaces. According to the source documentation, the template follows this structure:

生成[空间类型]设计效果图，功能定位为[用途]。
风格：[现代简约/工业/新中式]，材质：[木/石/金属/玻璃]。
空间结构：[开敞/分区]，动线：[主通道说明]。
光线：[自然采光/人工照明方案]，时间：[白天/夜景]。
输出：写实建筑空间渲染图。

JSON-Structured Template Format

The machine-friendly format enables automated pipelines and direct API integration. This structured approach is recommended when calling the backend endpoint defined in api/generate-image.js.

{
  "type": "Architectural Visualization",
  "space": {
    "type": "Modern Cabin Interior",
    "function": "Living room",
    "materials": "Exposed concrete, large floor-to-ceiling glass, warm timber accents"
  },
  "environment": "Nestled in a dense, snowy pine forest visible through the glass",
  "camera": {
    "angle": "Eye-level perspective, wide-angle lens",
    "lighting": "Golden hour, warm interior lights glowing, cool blue ambient light outside"
  },
  "render_quality": "Unreal Engine 5 style, hyper-realistic, 8k resolution, ray tracing"
}

Core Schema Fields for Architectural Visualization

Both template formats share eight critical fields that control the final render output. When constructing your payload for GPT-Image2, specify these parameters to ensure architectural accuracy:

• type: Defines the generation task. Use "Architectural Visualization" for consistency with the template family.
• space.type: Declares the architectural style and structure, such as "Modern Cabin Interior" or "Industrial Loft".
• space.function: Specifies the intended use, including "Living room", "Office", or "Lobby".
• space.materials: Lists dominant construction finishes like "Exposed concrete, large floor-to-ceiling glass, warm timber accents".
• environment: Provides narrative context for surrounding landscapes, for example, "Nestled in a dense, snowy pine forest visible through the glass".
• camera.angle: Locks the perspective to prevent distortion. The documentation recommends "Eye-level perspective, wide-angle lens" for standard architectural views.
• camera.lighting: Controls time-of-day and light temperature, such as "Golden hour, warm interior lights glowing, cool blue ambient light outside".
• render_quality: Specifies the target engine and resolution, typically "Unreal Engine 5 style, hyper-realistic, 8k resolution, ray tracing".

Implementation Examples

The repository provides reference implementations showing how to deploy these templates across different interfaces.

Node.js API Integration with JSON Payload

The api/generate-image.js endpoint accepts the JSON template as a POST body. Here is a production-ready implementation that structures the request correctly:

const fetch = require('node-fetch');

async function generateArchViz(payload) {
  const response = await fetch('https://api.your-gpt-image-service.com/generate', {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify(payload)
  });
  const result = await response.json();
  return result.imageUrl;
}

const archPayload = {
  type: "Architectural Visualization",
  space: {
    type: "Modern Cabin Interior",
    function: "Living room",
    materials: "Exposed concrete, large floor-to-ceiling glass, warm timber accents"
  },
  environment: "Nestled in a dense, snowy pine forest visible through the glass",
  camera: {
    angle: "Eye-level perspective, wide-angle lens",
    lighting": "Golden hour, warm interior lights glowing, cool blue ambient light outside"
  },
  render_quality: "Unreal Engine 5 style, hyper-realistic, 8k resolution, ray tracing"
};

generateArchViz(archPayload).then(url => console.log('Result:', url));

Chat Interface Prompting

For users interacting through the frontend defined in src/main.jsx, the plain-text template supports direct natural language input. Populate the bracketed placeholders with your specific project details:

生成现代简约风格的客厅设计效果图，功能定位为休闲交流空间。
材质：木/玻璃，空间结构：开放式，动线：从入口直接通向沙发区。
光线：自然采光，时间：白天。
输出：写实建筑空间渲染图。

Best Practices from the Source Code

The docs/templates.md file includes an "avoid-pitfalls" guide that highlights two critical techniques for achieving premium architectural renders:

Lock the camera perspective – Always specify an eye-level view in the camera.angle field to prevent distorted perspectives that break architectural realism.

Contrast warm and cold lighting – Combine warm interior lighting with cool exterior ambient light to create depth and premium visual quality. This technique is explicitly recommended in the template documentation for achieving cinematic architectural visualization.

Frontend and Backend Integration Points

The complete workflow for GPT-Image2 architecture templates spans three key files in the repository:

• docs/templates.md#tpl-architecture – Stores the master template definitions and schema documentation.
• src/main.jsx – Provides the React-based UI where users select or paste templates before submission.
• api/generate-image.js – Receives the structured payload and forwards it to the image generation service.

By customizing fields in the JSON payload or modifying the bracketed placeholders in the plain-text version, you can adapt the system to any architectural style, material palette, or lighting scenario required for concept work or client presentations.

Summary

• GPT-Image2 architecture templates provide dual-format options: plain-text for chat interfaces and JSON for API integration.
• The schema includes eight controlled fields covering space type, materials, environment, camera angles, and rendering quality.
• File docs/templates.md#tpl-architecture contains the complete template specifications and anti-pitfall guidelines.
• The src/main.jsx frontend and api/generate-image.js backend handle the template workflow from user input to image generation.
• Specific camera angles and contrasting lighting temperatures are required for high-quality architectural output.

Frequently Asked Questions

What file contains the architecture templates in GPT-Image2?

The complete architecture and space templates are defined in docs/templates.md under the #tpl-architecture section. This file contains both the plain-text Chinese template and the JSON schema, along with implementation guidelines for the src/main.jsx frontend and api/generate-image.js backend components.

How do I switch between plain-text and JSON formats?

Both formats share the same semantic schema, so you can convert between them by mapping the eight core fields. Use the plain-text template for direct chat input in src/main.jsx, and switch to the JSON structure when calling api/generate-image.js programmatically. The JSON format is required for automated pipelines that supply fields like space.materials and camera.lighting through variables.

What camera angle produces the best architectural renders?

According to the source documentation in docs/templates.md, you should specify "Eye-level perspective, wide-angle lens" in the camera.angle field. This prevents perspective distortion and maintains architectural accuracy, particularly for interior visualizations where vertical lines must remain true.

Can I specify Unreal Engine 5 rendering quality?

Yes. Set the render_quality field to "Unreal Engine 5 style, hyper-realistic, 8k resolution, ray tracing" to achieve game-engine aesthetic quality. This parameter works in both the JSON payload sent to api/generate-image.js and as a natural language instruction in the plain-text template's output specification line.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}