Consistent Social Media Notes with an Open-Source AI Skill

The Core Challenge

Creating engaging social media notes isn’t about generating individual images or writing captions—it’s about maintaining visual consistency across multiple posts. Without deliberate design systems, AI-generated content often suffers from:

• Inconsistent color schemes between covers and inner pages
• Disconnected character poses and proportions
• Erratic typography and layout patterns These inconsistencies make accounts appear disjointed, reducing brand recognition.

The Solution: Asset Boards

The open-source Omi3.0 skill introduces a “Visual Asset Board”—a design blueprint that enforces consistency:

1. Character Standards: Base illustrations of recurring IP characters
2. Color System: 5-6 color swatches with locked primary/secondary ratios
3. Typography Guide: Predefined font hierarchy for headings/body text
4. Layout Wireframes: Cover and inner page templates
5. UI Components: Reusable decorative elements

This isn’t a finished graphic, but a machine-readable style guide referenced during every image generation.

The Four-Step Workflow

Step 1: Asset Generator (One-Time Setup)

Input: Character reference + style description → Output: 16:9 asset board. Example parameters:

{
  "style": "minimalist tech",
  "ip_reference": "young female engineer cartoon",
  "color_mood": "vibrant primary with neutral accents"
}

Step 2: Note Director (LLM)

Generates a structured JSON storyboard specifying:

• Cover/content page types
• IP character roles (Hero/Guide modes)
• Exact layout intentions
• Color application rules

Step 3: Cover Generator

Uses the asset board + cover JSON to output 3:4 cover images (4K resolution). Hero Mode demands:

• IP occupies ≥40% of frame
• Dynamic poses interacting with text
• High-contrast title treatments

Step 4: Content Page Generator

Produces 3:4 inner pages (2K resolution) referencing the same asset board. Guide Mode requires:

• IP ≤15% size, often corner-positioned
• Subtle pointing/reading gestures
• Consistent margin and spacing

JSON Storyboard Structure

The director outputs machine-friendly instructions:

{
  "type": "cover",
  "content_layer": {
    "main_title": "5 Python Tricks You Aren't Using",
    "text_layout": "poster_style"
  },
  "visual_layer": {
    "ip_role": "Hero",
    "ip_instruction": "holding giant Python logo, excited expression"
  }
}

Customization Points

Modify without restructuring the pipeline:

1. Style Descriptions: Replace default aesthetic preferences
2. Page Count: Adjust from default 3-5 to 7 for detail-heavy topics
3. IP Visibility: Reduce cover presence to 30% for text-centric accounts
4. Content Rules: Add platform-specific prohibited terms

Platform Flexibility

The MIT-licensed skill works with any agent supporting image generation:

• OpenClaw
• Claude Code
• Kouzi (Bytedance’s platform)

For manual testing:

1. Extract Step 2’s JSON output
2. Feed it to Midjourney/Stable Diffusion with the asset board
3. Compare visual consistency across generations

Built-In Compliance

Prompt-level safeguards automatically reject:

• Absolute claims (“best”/“#1”)
• External link references
• VPN-related terminology
• Giveaway-focused pages

Implementation

1. Clone the repository:

git clone https://github.com/CzzzzzzJ/omi.git

2. Follow SKILL.md’s orchestration instructions
3. Modify prompts/ directory files for customization

(Figures and case details in this article are as reported in the source material and have not been independently verified.)

Series: AI Practice