asset-prompt-generator

name	asset-prompt-generator
description	Transforms narrative script data into cinematic prompt batches for FLUX/LTX.

Asset Prompt Generator Skill

Overview

A specialized skill for transforming narrative script data (J10) into highly technical, cinematic prompt batches optimized for FLUX and LTX video generation models.

Trigger Conditions

• When transitioning from Script Generation (D2) to Image/Video Generation (J8).
• When a project requires shot-by-shot visual breakdown based on a World Bible and Pilot Script.

Workflow

1. Input Ingestion: Load the parsed pilot_script.md (from J10 parser) and the world_bible.md.
2. Shot Decomposition: Use an LLM (via KimiBridge) to decompose each script scene into a sequence of distinct visual shots.
3. Cinematic Parameter Injection: Apply standardized technical descriptors for:
   • Camera Presets: wide, close_up, medium, tracking, drone.
   • Lighting Styles: cyberpunk neon, golden_hour, noir, etc.
4. Payload Construction: Generate a structured JSON list of payloads containing the base_prompt (flattened string) and metadata (camera, subject, lighting).

Technical Standards

• Prompt Structure: [Subject/Action/Environment]. [Camera Detail]. [Named light source and direction]. [Material texture and physical imperfections]. [Composition/focal priority].
• Specificity Floor: Each generated still prompt must include at least one concrete texture, one optical detail, one light-source detail, and one composition constraint.
• Generic Token Ban: Do not use masterpiece, best quality, ultra-high fidelity, or 8k as quality substitutes. Replace them with visible material and capture details.
• Output Format: A JSON-compatible list of dictionaries for direct injection into the ComfyUI API layer.

Pitfalls & Best Practices

• Avoid Generic Terms: Never use "cinematic" alone; always pair it with a specific descriptor (e.g., "cinematic scale," "cinematic lighting").
• Consistency Anchors: Always reference the World Bible's 'Consistency Anchors' to ensure character and environmental continuity across shots.
• Fallback Logic: If LLM decomposition fails, provide a default set of generic shots based on scene description to prevent pipeline stalls.

Related Skills

• genesis-engine-orchestration: Orchestrates the initial idea-to-script phase.
• comfyui-api-submission: The subsequent layer for executing these prompts.