21 Industrial-Grade Prompt Templates in awesome-gpt-image-2: A Complete Catalog

The awesome-gpt-image-2 repository provides 21 production-ready "2I" prompt templates, each embedding explicit pitfall-avoidance constraints to ensure consistent, high-fidelity output from GPT-Image-2 models.

The freestylefly/awesome-gpt-image-2 project ships a standardized framework for generating professional visuals through structured prompting. These 21 industrial-grade prompt templates—internally designated as the 2I series—are documented in [docs/templates.md](https://github.com/freestylefly/awesome-gpt-image-2/blob/main/docs/templates.md) and provide battle-tested schemas for everything from mobile UI mockups to architectural renderings.

What Makes a Template "Industrial-Grade"

Unlike generic prompts, every 2I template includes a mandatory "avoid-pitfalls" block that explicitly defines negative constraints. This architecture prevents common model failures such as garbled text in screenshots, perspective distortion in architecture, or anachronistic elements in historical scenes.

The templates are stored in three key locations:

• docs/templates.md – Human-readable master documentation with full prompt skeletons
• data/style-library.json – Machine-readable JSON for programmatic agent consumption
• agents/skills/gpt-image-2-style-library/SKILL.md – Skill declaration for agent integration

The Complete 2I Template Catalog

The 21 templates are organized by visual domain. Below are the documented identifiers and their specific constraints as implemented in the source files.

UI and Social Media Templates

tpl-ui – Generates high-fidelity interface mockups for specific platforms. The skeleton requires defining the platform (iOS/Android), aspect ratio (9:16 or 16:9), and visual hierarchy. Pitfall avoidance enforces readable typography and prohibits vague layout instructions.

tpl-screenshot – Creates social media content screenshots with locked color modes (dark/light). The template mandates accurate Chinese text display and explicitly forbids character encoding errors or "garbled text" common in generated screenshots.

tpl-live – Produces live-stream interface overlays. Constraints include clear UI layer definitions to prevent obstruction of the host, plus readable bullet-comment (danmu) placement.

Marketing and Infographic Templates

tpl-infographic – Structures data visualization with 3–5 content modules and specific chart types. The pitfall-avoidance rule limits total text volume to maintain readability.

tpl-poster – Designs event, product, or movie posters. Requires locked subject matter, dominant color palette, and composition style to prevent generic results.

tpl-campaign – Commercial campaign visuals with strict structural definitions (single image, triptych, or data doodle). Prevents cluttered collages by enforcing clear visual hierarchy and unified color tones.

Typography and Brand Identity

tpl-typography – Premium conceptual typography posters. The constraint system is rigid: the title must appear as a complete, readable main visual, with no additional text or icons permitted.

tpl-signature – Multi-style signature design system. Generates vertical (9:16) signature selection posters with unified spacing and margins. Explicitly forbids font collages and chaotic color schemes.

tpl-brand – Comprehensive brand visual systems including logo directions, auxiliary graphics, and color palettes. Requires defined brand keywords and target audience, with mandatory white backgrounds for consistency.

Product and Commercial Photography

tpl-product – E-commerce main images. The template enforces material detail capture, specific lighting setups, and limits promotional copy to prevent cluttered commercial outputs.

Photorealistic and Artistic Rendering

tpl-architecture – Photorealistic space rendering. Uses camera parameters to control perspective and lighting contrast, with explicit rules against perspective distortion.

tpl-photo – High-realism photography simulation. Requires specifying camera parameters (focal length, aperture) and lighting conditions. The pitfall-avoidance block strictly excludes CG or illustration aesthetics.

tpl-illustration – Stylized illustrations for covers or social media. Locks brushwork style and color palette. Notably forbids direct usage of master artist names to avoid copyright issues while maintaining style consistency.

tpl-character – Character design sheets. Defines appearance, personality, and world-building elements with clear pose and expression specifications to prevent multi-character confusion.

tpl-scene – Narrative scene generation. Requires explicit event definition, protagonist placement, and cinematic lens language (shot types) to create dramatic tension.

tpl-history – Historical and ancient-style imagery. Excludes all modern elements and mandates accurate cultural details for costumes and architecture.

Publishing and Document Layouts

tpl-document – Publication layout design. Controls column count, margins, and white space. Prevents "wall of text" layouts by enforcing modular content blocks.

tpl-other – Generic fallback template for undefined use cases. Requires explicit task goals and constraints, outputting both a primary scheme and an alternative (A/B testing).

Implementation and Agent Integration

To use these templates programmatically, reference the template ID when calling the style library skill.

# Example: Referencing a template via the skill system

skill_request = {
    "skill": "gpt-image-2-style-library",
    "template_id": "tpl-ui",
    "params": {
        "platform": "iOS",
        "product": "Fitness App",
        "aspect_ratio": "9:16"
    }
}

The data/style-library.json file provides the schema definitions for each template, allowing agents to validate parameters before sending requests to the image generation API. According to [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), agents select templates by ID and populate the bracketed variables in the prompt skeletons.

JSON vs. Plain Text Formats

Templates support two input modes. Plain text uses fill-in-the-blank structures:

为[产品类型]生成一张[平台]界面图。核心功能：[…]视觉风格：[风格]，主色[颜色]，布局[…] 输出：高保真UI截图，文字清晰可读，比例[9:16/16:9]。

For agent automation, JSON schemas provide structured control:

{
  "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

• 21 industrial-grade templates in the 2I series cover UI, marketing, typography, products, photography, and publishing domains.
• Pitfall-avoidance rules are embedded in every template to prevent common generation errors like garbled text or distorted perspective.
• Three file locations define the system: docs/templates.md for humans, data/style-library.json for machines, and SKILL.md for agent integration.
• Dual-format support allows both manual plain-text filling and programmatic JSON consumption.
• Strict constraints (e.g., mandatory aspect ratios, color locks, and negative prompts) ensure production-ready consistency.

Frequently Asked Questions

What does "2I" mean in the awesome-gpt-image-2 templates?

2I is the internal designation for this collection of 21 industrial-grade prompt templates. The abbreviation signifies the project's focus on "industrial" reliability and pitfall prevention, distinguishing these structured prompts from ad-hoc user queries.

Can I modify the pitfall-avoidance rules in the templates?

Yes. The master definitions reside in docs/templates.md and data/style-library.json. You can fork the repository and edit the constraint blocks to match specific brand guidelines or technical requirements, provided you maintain the JSON schema structure for agent compatibility.

How do agents select the correct template automatically?

Agents use the skill declared in agents/skills/gpt-image-2-style-library/SKILL.md to map user intent to template IDs. The skill references data/style-library.json to validate that required fields (like platform for tpl-ui or theme for tpl-photo) are present before generation.

Why do some templates specify camera parameters?

Templates like tpl-photo and tpl-architecture include camera parameters (focal length, aperture, lighting angles) to force photorealistic rendering characteristics. According to the source code analysis, explicitly defining these parameters prevents the model from defaulting to "CG" or illustration aesthetics, ensuring outputs mimic real-world optical physics.

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