Prompt Master

The Problem It Solves

When using AI generation tools like Midjourney, Stable Diffusion, or Sora, users encounter several key challenges:
- High barrier to prompt writing: These tools rely on precise textual descriptions, but average users struggle with professional terminology and structured phrasing, leading to subpar results.
- Inconsistent platform syntax: Different tools (e.g., MJ uses parameter suffixes, SD uses bracket weights) have varying prompt formats, causing errors when switching platforms.
- Safety and compliance risks: Generated content may inadvertently infringe on portrait rights, copyrights, or contain sensitive information, with no automatic validation.
- Low iteration efficiency: Adjusting prompts often requires manual restructuring, making fine-tuning time-consuming.

This skill automates the creation of high-quality, platform-compliant prompts through intelligent routing and structured workflows, lowering the entry barrier.

How the Skill Works

The core is an intent-driven, multi-stage processing pipeline:

1. Intent Routing and Requirement Research
   - Based on user input (keywords or uploaded files), it automatically identifies task types (mode_image, mode_video, mode_teach, etc.).
   - Research collects key fields: q_type (generation type), q_theme (subject content), q_style (style preference), q_platform (target platform) to ensure precise understanding.

2. Safety Validation and Feasibility Checks
   - Pre-generation rule checks include:

   • check_portrait: Refuses to generate real human portraits to avoid infringement.
   • check_sensitive: Filters violent, pornographic, or sensitive content.
   • check_copyright: Replaces copyrighted characters with generic descriptions.
   • Physically impossible descriptions trigger warnings but don't block generation.

3. Field Decomposition and Smart Filling
   - Activates different field tiers (P0 core, P1 extended, etc.) based on input detail level (input_level).
   - Uses smart defaults (e.g., F-ENV defaults to "simple background, soft lighting") to fill gaps and trims to prevent redundancy.

4. Platform Syntax Adaptation
   - Converts prompt formats per target platform:

   • Midjourney: Natural language with suffix parameters (e.g., --ar 16:9).
   • SD: Supports bracket weights (e.g., (keyword:1.5)).
   • Domestic platforms (e.g., Doubao, Jimeng): Translates to natural language, removing weight symbols.

5. Output Generation and Self-Check
   - Outputs positive and negative prompts in a unified format, e.g.:
   Positive prompt: a serene landscape, soft lighting, 4k detailed Negative prompt: blurry, low quality, text
   - Performs completeness checks (non-empty fields, correct syntax) and quality checks (e.g., keyword count within range).

Additionally, the skill supports image-to-image, video generation, and reverse engineering: uploading a reference image allows prompt extraction or redraw based on parameters; the decompile feature deconstructs elements of existing images/videos (e.g., F-COMP composition, F-LIGHT lighting) to aid iteration.

Applicable Boundaries and Considerations

• Input Dependency: Output quality heavily relies on the user's initial description; vague inputs (under 10 characters) may trigger default fills, requiring more details.
• Platform Limitations: Some platforms (e.g., DALL-E) don't support weight syntax; the skill automatically removes such symbols, but users should be aware of final prompt differences.
• Safety Constraints: Strictly refuses real human or sensitive content generation; copyright issues may be mitigated via description substitution but aren't absolutely compliant.
• Teaching Mode: Entering "teach me" provides style guides and examples but doesn't guarantee identical generation results.
• Session Context: Research results are valid only within a single session; restarting requires re-collecting information.

This skill streamlines prompt writing through structured workflows, but final generation effects still depend on AI model capabilities and user collaborative adjustments.