GPT-Image2 Prompt Formats: Plain-Text vs. JSON Templates Explained

The GPT-Image2 template library supports two distinct prompt formats—a plain-text template for human-readable instructions and a structured JSON template for programmatic validation and agent automation.

The freestylefly/awesome-gpt-image-2 repository defines a comprehensive prompt engineering system for generating high-fidelity images with GPT-Image2. Understanding the available prompt formats is essential for choosing the right workflow, whether you are manually testing prompts in the web UI or building automated agents that generate images at scale.

Core Prompt Format Types

The library standardizes on two representations for every domain-specific template. According to the source code in docs/templates.md, each category—including UI, infographic, poster, product, brand, architecture, photography, illustration, character, scene, history, and document—provides both formats.

Plain-Text Human-Readable Templates

This format presents a free-form block of natural language instructions following conventional "Generate a..." phrasing. As implemented in lines 16-22 of [docs/templates.md](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/docs/templates.md), plain-text templates list required fields such as product type, platform, style, layout, and aspect ratio in a conversational structure. This format is ideal for quick one-off prompts, manual testing, and hand-offs to human editors who need to copy-and-paste directly into the web UI gallery.

JSON Structured Templates

The JSON format expresses each prompt component as typed key-value pairs with explicit constraints. Lines 24-46 of docs/templates.md demonstrate this schema, which mirrors the plain-text fields but adds nesting, validation rules, and deterministic structure. The JSON version is recommended for automated agents because it can be parsed and validated before sending to the model, preventing ambiguous wording that might confuse the model while allowing the gpt-image-2-style-library skill to fill placeholders programmatically.

Source Files Defining the Prompt System

Several key files govern how these formats are stored and accessed:

• [docs/templates.md](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/docs/templates.md): Master document containing every template in both plain-text and JSON versions across all categories.
• [data/style-library.json](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/data/style-library.json): Machine-readable catalogue that agents query to retrieve JSON schemas programmatically.
• [agents/skills/gpt-image-2-style-library/references/style-library.md](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/agents/skills/gpt-image-2-style-library/references/style-library.md): Index linking each category to its concrete prompt format.
• [README.md](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/README.md): Provides workflow guidance on selecting the appropriate format for your use case.

Code Implementation Examples

Below are minimal implementations for the UI category, demonstrating both formats side-by-side.

Plain-text version suitable for direct UI input:

// Plain-text prompt (copy-and-paste)
const plainPrompt = `
为[产品类型]生成一张[平台，如 iOS/Android/Web]界面图。
核心功能：[功能点A]、[功能点B]、[功能点C]。
视觉风格：[极简/科技/拟物]，主色[颜色]，强调色[颜色]。
布局：[顶部导航/双栏/卡片流]，信息层级清晰，留白充足。
输出：高保真UI截图，文字清晰可读，比例[9:16/16:9]。
`;

Structured JSON version for agent automation:

{
  "type": "UI Screenshot",
  "platform": "iOS",
  "product": "Fitness App",
  "layout": "Card-based feed with bottom tab bar",
  "style": {
    "theme": "Dark Mode",
    "primary_color": "Neon Green",
    "typography": "Clean sans-serif"
  },
  "content": {
    "header": "Today's Activity",
    "cards": [
      {"title": "Running", "data": "5.2 km", "button": "Start"},
      {"title": "Calories", "data": "340 kcal"}
    ]
  },
  "constraints": "High fidelity, readable text, 9:16 aspect ratio"
}

You can invoke either format via CLI:

# Using the structured JSON version

gpt-image2 generate --json file.json

# Using the plain-text version

gpt-image2 generate --prompt "$(cat plain.txt)"

Summary

• The GPT-Image2 library provides two standardized prompt formats: plain-text for human editing and JSON for machine parsing.
• Plain-text templates appear in docs/templates.md (lines 16-22) and support copy-and-paste workflows.
• JSON templates appear in docs/templates.md (lines 24-46) and enable validation, constraint enforcement, and agent automation.
• The [data/style-library.json](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/data/style-library.json) file serves as the machine-readable index for all template schemas.
• Both formats cover identical domains: UI, infographic, poster, product, brand, architecture, photography, illustration, character, scene, history, and document.

Frequently Asked Questions

What is the difference between plain-text and JSON prompts in GPT-Image2?

Plain-text prompts present instructions as natural language paragraphs suitable for direct pasting into the web UI. JSON prompts structure the same requirements as typed key-value pairs, enabling programmatic validation and precise constraint enforcement before generation.

How do I programmatically validate a GPT-Image2 prompt before sending it?

Parse the JSON template from data/style-library.json against your input object to verify required fields like type, platform, and constraints are present and correctly typed. The JSON schema prevents ambiguous parameters that might degrade image quality.

Which prompt format should I use for automated agent workflows?

Use the JSON structured template when building agents. This format allows the gpt-image-2-style-library skill to fill placeholders algorithmically, enforce aspect ratios and color constraints, and integrate with validation pipelines.

Where can I find the complete list of available templates?

The master index resides in [docs/templates.md](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/docs/templates.md), with a machine-readable catalogue in [data/style-library.json](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/data/style-library.json). The [README.md](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/README.md) provides quick links to both resources.

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