Anti-Pitfall Guides for Awesome-GPT-Image-2 Templates: Avoiding Common AI Image Generation Mistakes

Awesome-GPT-Image-2 templates include domain-specific "avoid-pitfall guides" (避坑指南) that enforce explicit constraints, negative directives, and sequence hints to eliminate hallucinations, unreadable text, and style mismatches in AI-generated images.

The freestylefly/awesome-gpt-image-2 repository provides more than 20 industrial-grade prompt templates spanning UI design, infographics, photography, and character art. Each template embeds an anti-pitfall guide that instructs the model exactly what to lock, enforce, or exclude. These guides follow a strict three-part architecture: explicit constraints (concrete attributes like aspect ratio), negative directives (banned elements like "no gibberish text"), and sequence hints (order of operations such as locking layout before style).

What Are Anti-Pitfall Guides in Awesome-GPT-Image-2?

Anti-pitfall guides are structured rule sets embedded in docs/templates.md that accompany every visual domain template. According to the source code in shared/apimart.js, the system parses these markdown guides and concatenates them with base prompts before sending requests to the image generation API.

The guides serve three critical functions:

1. Explicit Constraints – Require concrete, measurable attributes (platform, aspect ratio, color hex codes, layout grid).
2. Negative Directives – Hard prohibitions such as "no modern elements" or "don't fill the canvas with paragraphs."
3. Sequence Hints – Mandatory ordering advice (e.g., "lock the UI layout first, then specify visual style").

Domain-Specific Anti-Pitfall Strategies

UI & Screenshots

The UI template in docs/templates.md demands precise platform specification to avoid generic interfaces.

• Always specify platform (iOS, Android, Web), aspect ratio, and layout type.
• Force text readability and lock exact Chinese characters to prevent garbled output.
• Distinguish platform-specific cues (e.g., Twitter's blue-check vs. Douyin's music disc icon).

Infographics

The infographic guide controls information density to prevent cluttered, unreadable visuals.

• Limit module count and explicitly declare chart type (flowchart, bar graph, pie chart).
• Keep copy short; prohibit paragraph overload on the canvas.
• Enforce clear visual hierarchy through size and color constraints.

Posters & Layouts

Poster templates enforce typographic dominance and prohibit lazy design patterns.

• The title must be 100% correct and occupy the top of the visual hierarchy.
• Ban moodboards, multi-poster grids, and generic 3D lettering.
• Require specific alignment and margin rules before generating content.

Product & E-Commerce

Product photography templates emphasize material realism to avoid "stall-goods" aesthetics.

• Emphasize material and lighting specifications; without them, renders look artificial.
• Restrict promotional copy to 1-2 lines maximum.
• For beauty recommendations, require skin-tone analysis before color palette mapping.

Brand & Identity

Brand templates start with semantic keywords rather than visual descriptions.

• Begin with brand keywords before requesting visual output.
• Enforce pure white backgrounds for easy cut-out and asset reuse.
• Provide explicit "don't-do" rules for color misuse and logo placement.

Architecture & Space

Architectural guides lock perspective to prevent distortion.

• Lock camera perspective (eye-level, aerial, worm's-eye) to eliminate lens distortion.
• Use cold-warm lighting contrast to create premium spatial depth.
• Prohibit floating structures or physically impossible geometry.

Photography & Realism

Photography templates add intentional imperfections to avoid plastic CGI looks.

• Add realistic imperfections (skin pores, film grain, dust) to prevent synthetic appearances.
• Use explicit photographic parameters (f-stop, focal length, ISO) instead of vague adjectives like "professional" or "high quality."

Illustration & Art

Illustration guides require technical specificity over vague artistic references.

• Pin down brush style, medium, and color palette explicitly.
• Avoid generic artist names; instead describe signature techniques (e.g., "wet-on-wet watercolor blending" rather than "in the style of Monet").

Character & Avatar

Character design templates require exhaustive attribute lists to prevent generic faces.

• Detail every visual attribute: age, hairstyle, clothing texture, and accessories.
• Lock pose and emotion explicitly.
• For action-sheet references, enforce a fixed grid layout and prohibit background clutter.

Scene & Narrative

Scene templates use kinetic language to generate motion.

• Include a verb-driven event ("exploding," "collapsing," "dissolving") to imply motion.
• Specify camera language (low-angle, Dutch angle, tracking shot) to shape dramatic tension.

Historical / Period

Historical templates enforce temporal purity.

• State the exact dynasty or era (e.g., "Tang Dynasty, 8th century").
• Add "no modern elements" as a hard negative constraint to prevent anachronisms.

Document & Publication

Document templates define structure before content.

• Define columns, margins, and hierarchy before requesting visual elements.
• Use placeholder text ("Simulated text blocks") rather than full sentences to prevent distraction.

Other Applications

General-purpose templates follow a constraint-first methodology.

• Begin with a concise task statement followed by constraint lists.
• Request both a primary and alternative solution for rapid A/B testing.

How the Anti-Pitfall System Works

The implementation spans four key files: docs/templates.md, src/main.jsx, shared/apimart.js, and src/apimartClient.js.

Template Selection – The front-end in src/main.jsx renders a dropdown of categories. When a user selects "UI Screenshot" or "Product Photography," the component retrieves the corresponding markdown section.

Prompt Assembly – The shared/apimart.js library parses the markdown, extracts the "避坑指南" section, and concatenates it with the base prompt. This ensures every generation request includes the domain-specific constraints.

JSON Mode – For programmatic access, the system accepts structured JSON objects that encode constraints directly. The src/apimartClient.js validates these objects against the anti-pitfall schema before forwarding to the /api/generate-image endpoint.

Result Validation – The client validates that required keywords (e.g., "文字绝对可读" for text readability) appear in the generated image metadata. If constraints are violated, the system retries with reinforced negative directives.

Practical Code Examples

UI Screenshot (JSON Mode)

{
  "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. 文字必须准确显示指定的中文，禁止乱码"
}

The constraints field directly embeds the UI anti-pitfall rule prohibiting garbled Chinese characters.

Infographic (Plain Text Prompt)

生成 老年人日常健康管理指南 信息图，目标读者为 65-75 岁退休人士。
结构：标题区 + 3 个模块，每模块含图标、短标题、1-2 句说明。
图表类型：流程图。
风格：专业报告，主色 蓝色，背景 浅色。
约束：信息层级清晰、可读性高的中文信息图。
不要把大量正文塞进画面，保持模块数量固定。

The final two lines implement the infographic anti-pitfall directives against text overload and module proliferation.

Product Hero Shot (Text Prompt)

生成 噪声消除耳机 电商主图。
场景：极简工作室，背景 灰色渐变。
灯光：软箱顶光 + 边缘背光。
材质：磨砂金属，标明 “NEW” badge。
约束：商业摄影质量，超高真实感。
不要把促销文字占满全屏，保持主体清晰。

The closing constraint "不要把促销文字占满全屏" originates from the product anti-pitfall guide's rule against promotional text crowding.

Summary

• Anti-pitfall guides (避坑指南) in docs/templates.md provide domain-specific guardrails for 20+ visual categories.
• Each guide follows a three-layer architecture: explicit constraints, negative directives, and sequence hints.
• The system implementation spans src/main.jsx (UI), shared/apimart.js (parsing), and src/apimartClient.js (API communication).
• JSON Mode allows programmatic constraint encoding, while plain-text prompts embed negative directives directly.
• Specific prohibitions include: no garbled text (UI), no paragraph overload (infographics), no modern elements (historical scenes), and no full-screen promotional text (product shots).

Frequently Asked Questions

How do I prevent unreadable text in AI-generated UI screenshots?

Specify the exact Chinese characters required and add the constraint "文字绝对可读" (text must be absolutely readable). According to the UI template in docs/templates.md, you must explicitly lock the platform (iOS/Android), aspect ratio, and distinguish platform-specific UI cues like Twitter's blue-check marks to avoid generic interface hallucinations.

What is the most common pitfall when generating product photography?

The product template identifies "stall-goods" aesthetics—images that look like low-quality market photos—as the primary failure mode. To avoid this, you must explicitly specify material texture (e.g., "brushed aluminum") and lighting setup (e.g., "softbox top light with rim backlight") rather than using vague terms like "professional lighting."

Can I use these anti-pitfall guides with non-Chinese prompts?

Yes. While the repository's docs/templates.md uses Chinese terminology (避坑指南), the architectural principles apply to any language. The shared/apimart.js parser extracts constraints regardless of language, and the JSON Mode examples demonstrate how to encode these rules in English or other languages for the underlying diffusion model.

Where does the system validate that anti-pitfall rules were followed?

The validation occurs in the client layer (src/apimartClient.js) and through metadata inspection. The system checks that required keywords from the anti-pitfall guides appear in the generated image metadata. If validation fails—such as missing text readability tags for UI screenshots—the client automatically retries with reinforced negative directives to ensure compliance.

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