21 Industrial-Grade Prompt Templates in awesome-gpt-image-2: The Complete 2I Series Guide

The awesome-gpt-image-2 repository provides 21 production-ready "2I" prompt templates with built-in pitfall-avoidance constraints, enabling consistent, high-fidelity image generation across UI design, marketing campaigns, photography, and illustration workflows.

The awesome-gpt-image-2 project is a curated toolkit designed to stabilize output quality for GPT-Image-2 agents. At its core sits a collection of 21 industrial-grade prompt templates (the 2I series), each engineered with explicit "avoid-pitfalls" guardrails that eliminate ambiguity and enforce reproducible visual standards. These templates are formally documented in [docs/templates.md](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/docs/templates.md) and exposed programmatically via the style-library skill system.

Anatomy of an Industrial-Grade Template

Industrial-grade status in the 2I series requires three structural components. Each template includes a prompt skeleton with variable placeholders (e.g., [产品类型], [主题]), an avoid-pitfalls block specifying exclusions (e.g., "禁止乱码", "避免透视失真"), and a constraint layer defining aspect ratios, color systems, or typographic rules. According to the source code in docs/templates.md, this architecture ensures that GPT-Image-2 receives unambiguous instructions rather than open-ended creative briefs.

The templates support dual consumption modes: plain text for manual copy-paste workflows and structured JSON for programmatic agent calls. The JSON schemas are codified in [data/style-library.json](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/data/style-library.json), while the agent integration logic resides in [agents/skills/gpt-image-2-style-library/SKILL.md](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/agents/skills/gpt-image-2-style-library/SKILL.md).

The 21 Industrial-Grade Prompt Templates

The 2I series spans six visual design domains. While docs/templates.md enumerates the full set of 21 templates, the following 18 core templates define the repository's coverage, with three additional specialized variants completing the industrial-grade catalog.

Interface and Social Media

These three templates enforce platform-specific UI constraints and legibility standards.

• tpl-ui: Generates high-fidelity app interface screenshots. The skeleton requires explicit platform declaration (iOS/Android), aspect ratio locking (9:16 or 16:9), and mandatory readable typography. Pitfall avoidance mandates layout hierarchy and prohibits ambiguous color instructions.

• tpl-screenshot: Produces social-media content screenshots with dark/light mode specifications. Critical constraints include accurate Chinese character rendering ("禁止乱码") and clear information layering for account details and engagement metrics.

• tpl-live: Creates live-stream interface mockups with UI overlay definitions. The template locks streamer positioning, prevents host obstruction by interface elements, and enforces readable bullet-comment rendering.

Marketing and Commercial Visuals

Four templates govern e-commerce and campaign imagery with strict commercial constraints.

• tpl-product: Designs e-commerce main images with detailed material and lighting specifications. Limits promotional copy quantity to prevent clutter while enforcing platform-ready resolution standards.

• tpl-poster: Generates event, product, or movie posters. Requires locked subject definition, color system declaration, and high-resolution output suitable for social distribution. Prohibits text堆砌 (text堆砌/text cluttering).

• tpl-campaign: Handles commercial campaign visuals for sports or fitness categories. Enforces structural selection (single image, triptych, or data-doodle format) before generation, preventing杂乱拼贴 (messy collages).

• tpl-infographic: Creates data visualization with 3-5 module structures. Specifies chart types and limits text volume to maintain hierarchical clarity and readability in Chinese layouts.

Brand Identity and Typography

Three templates focus on systematic brand representation and typographic precision.

• tpl-brand: Develops complete visual identity systems including logo directions, auxiliary graphics, and primary/secondary color definitions. Forces pure white background application and locks brand keyword alignment.

• tpl-typography: Produces conceptual typography posters for exact title specifications. Critical constraints mandate complete title readability as the primary visual element while prohibiting additional text or icon elements.

• tpl-signature: Generates multi-style signature selection posters in 9:16 vertical format. Enforces uniform spacing, whitespace balance, and prohibits font collage or杂乱颜色 (messy colors).

Creative Arts and Photography

Four templates distinguish between photographic realism and illustrative styles.

• tpl-photo: Commands photorealistic generation using explicit camera parameters (focal length, aperture settings). Explicitly avoids CG or illustration styles to maintain photographic authenticity.

• tpl-illustration: Creates cover-quality illustrations with locked brushwork and style specifications. Prohibits direct invocation of master artist names to prevent copyright issues while controlling line quality and color palette.

• tpl-character: Produces character design sheets with defined posture and expression constraints. Prevents multi-character confusion by enforcing single-subject focus with标志性元素 (iconic elements).

• tpl-scene: Generates narrative scene concept art with cinematic language. Requires explicit event/conflict definition and uses镜头语言 (shot language) to强化戏剧 (strengthen drama).

Spatial Design and Cultural Visuals

Two specialized templates handle architectural and historical accuracy.

• tpl-architecture: Renders architectural spaces with controlled perspective and light contrast. Prohibits透视失真 (perspective distortion) through explicit viewpoint and material specifications.

• tpl-history: Produces historically accurate古风 (ancient/traditional Chinese) imagery. Excludes modern elements while enforcing detailed costume and architectural accuracy for specific dynastic periods.

Document Layout and General Purpose

These templates handle publishing layouts and edge cases.

• tpl-document: Creates publication layouts with locked column counts and whitespace requirements. Prevents全文字 (all-text) layouts by mandating visual module integration.

• tpl-other: Serves as a general-purpose template for undefined scenarios. Requires explicit goal and constraint declaration, outputting both a primary and alternative solution for A/B testing workflows.

Implementation and Agent Integration

The 2I series integrates with agent systems through a declared skill interface. According to agents/skills/gpt-image-2-style-library/SKILL.md, agents invoke templates by ID (e.g., tpl-ui) and retrieve schema definitions from data/style-library.json. This JSON file maps each template ID to its parameter structure, enabling automated prompt assembly without manual string concatenation.

The pitfall-avoidance system operates through negation constraints in the JSON schema. Each template object includes an exclusions array listing prohibited outputs (e.g., "避免CG感" for photography templates), which the agent checks against during post-processing validation.

Usage Examples

The following examples demonstrate the dual-format usage for interface generation, taken directly from docs/templates.md:

Plain text invocation for manual workflows:

为社交媒体应用生成一张iOS界面图。
核心功能：好友列表、即时消息、推送通知。
视觉风格：极简，主色蓝色，强调色橙色。
布局：双栏，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。

Structured JSON for agent consumption:

{
  "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"
}

Summary

• awesome-gpt-image-2 defines 21 industrial-grade prompt templates (the 2I series) documented in docs/templates.md.
• Each template includes a prompt skeleton, pitfall-avoidance constraints, and output specifications to ensure reproducible quality.
• Templates are categorized across six domains: Interface, Marketing, Branding, Creative Arts, Spatial/Cultural, and General Purpose.
• Agent integration is supported via data/style-library.json and the gpt-image-2-style-library skill defined in SKILL.md.
• Dual-format support enables both manual text copying and programmatic JSON invocation.

Frequently Asked Questions

What does "2I" stand for in the template series?

The "2I" designation refers to the "Industrial-Grade" standard applied to GPT-Image-2 workflows. It signifies that each template includes built-in "avoid-pitfalls" (避坑) guardrails—explicit negative constraints that prevent common generation errors like text garbling, perspective distortion, or stylistic drift.

How do I choose between the text and JSON template formats?

Use the text format (docs/templates.md) for manual copy-paste generation in chat interfaces where you fill bracketed placeholders like [产品类型] manually. Use the JSON format (data/style-library.json) when building automated agents that programmatically assemble prompts via the style-library skill, allowing structured parameter injection and validation.

Can I modify the 21 templates or add custom variants?

Yes. The repository structure in docs/templates.md uses a consistent schema that you can extend. To add a custom template, define a new tpl-[name] ID, include a prompt skeleton with variable placeholders, and specify an avoid-pitfalls block. For agent integration, add the corresponding schema entry to data/style-library.json and update SKILL.md to register the new template ID.

Are these templates specific to Chinese-language generation?

While the prompt skeletons in docs/templates.md use Chinese placeholders and cultural contexts (e.g.,古风 for historical styles), the underlying constraint architecture works for any language. The tpl-typography template explicitly includes English instructions ("Create ONE finished premium conceptual typography poster"), and the JSON schemas use English keys, making the 2I series adaptable for multilingual GPT-Image-2 workflows.

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