visual-asset-prompt-generator

name	visual-asset-prompt-generator
description	Reads brand documentation (visual identity, product description, positioning) and generates optimized prompts for image generation (Nano Banana/Flux via fal.ai) and 3D model generation (Hunyuan3D). Use when generating visual assets from brand docs.

Visual Asset Prompt Generator

Extracts brand context from documentation and generates structured prompts for AI image and 3D model generation, tailored to the output target (landing page, pitch deck, website).

When to Use

• User wants to generate visual assets (images, 3D models) for a brand
• User says "generate visuals", "create assets", "brand images", "3D models from brand"
• As Step 1 of the brand-visual-pipeline

Input Variables

• [BRAND_NAME] — Name of the brand (matches brands/{name}/ directory)
• [BRAND_DOCS_DIR] — Path to brand docs (default: brands/{BRAND_NAME}/)
• [OUTPUT_CONTEXT] — Target: landing-page | pitch-deck | website | social
• [ASSET_TYPES] — What to generate: hero-image | product-shot | 3d-model | texture | all
• [PRODUCT_FOCUS] — Optional: specific product/feature to emphasize

The Protocol

Phase 1: Brand Context Extraction

Read the following brand docs (in order of priority):

1. Visual Identity (06-visual-identity.md or visual-identity-guide.md)

   • Extract: color palette (hex codes), mood keywords, imagery style, illustration style
   • Extract: what to AVOID (visual positioning "Avoid" column)
   • Extract: graphic elements (patterns, textures, borders)

2. Product Description (04-detailed-product-description.md)

   • Extract: product features, physical characteristics, materials
   • Extract: core offerings and what the product IS

3. Product Positioning (02-product-positioning.md)

   • Extract: market position, brand essence, target perception

4. Logo Prompts (12-logo-prompts.md, if exists)

   • Extract: visual keywords, symbolic concepts, brand essence visuals

5. MDS (05-messaging-direction-summary.md)

   • Extract: pitch, USP, emotional drivers

Phase 2: Build Brand Context Object

Compile extracted data into a structured context:

{
  "brand": "[BRAND_NAME]",
  "colors": {
    "primary": { "name": "...", "hex": "#..." },
    "secondary": { "name": "...", "hex": "#..." },
    "accent": { "name": "...", "hex": "#..." }
  },
  "mood_keywords": ["keyword1", "keyword2", ...],
  "imagery_style": "description of photography/illustration style",
  "avoid": ["thing1", "thing2", ...],
  "product_essence": "one-line product description",
  "materials": ["material1", "material2"],
  "graphic_elements": "patterns, textures, etc."
}

Phase 3: Generate Prompts by Asset Type

For Hero Images (fal.ai / Nano Banana / Flux):

• Compose a detailed scene description incorporating brand mood and colors
• Include lighting style from imagery guidelines
• Add negative prompt from avoid list
• Tailor composition to output context:
  • landing-page: Wide, atmospheric, emotional — 16:9 or 21:9 aspect ratio
  • pitch-deck: Clean, professional, product-focused — 16:9
  • website: Versatile, multiple crops possible — 3:2 or 4:3
  • social: Square or 4:5, high impact, minimal

For Product Shots (fal.ai / Nano Banana / Flux):

• Focus on product features and materials
• Use brand color palette for background/environment
• Clean studio-style or contextual lifestyle based on brand guidelines
• Include texture and material details

For 3D Models (Hunyuan3D):

• Keep text_prompt under 200 characters (Hunyuan3D limit)
• Focus on object description, not scene
• Include material and finish (matte, glossy, textured)
• Calculate bbox_condition ratios from product dimensions
• Generate as English-only prompt

For Textures (PolyHaven search terms):

• Generate search category keywords from brand materials
• Match brand aesthetic (e.g., "wood" for craft, "stone" for architectural)

Phase 4: Context-Specific Optimization

Apply output-target modifications:

Context	Emphasis	Style	Format
landing-page	Emotional, atmospheric	Lifestyle + product	16:9, 1920x1080
pitch-deck	Professional, clear	Product on clean BG	16:9, 1920x1080
website	Versatile, consistent	Multiple angles	3:2, 1200x800
social	High impact, bold	Minimal, eye-catching	1:1, 1080x1080

Output Format

Generate a JSON file at brands/{BRAND_NAME}/visual-prompts.json:

{
  "brand": "[BRAND_NAME]",
  "generated_at": "ISO_DATE",
  "output_context": "[OUTPUT_CONTEXT]",
  "brand_context": {
    "colors": { ... },
    "mood_keywords": [...],
    "avoid": [...],
    "imagery_style": "..."
  },
  "prompts": [
    {
      "asset_id": "hero-001",
      "type": "hero-image",
      "service": "fal-flux",
      "prompt": "detailed generation prompt...",
      "negative_prompt": "things to avoid...",
      "parameters": {
        "width": 1920,
        "height": 1080,
        "guidance_scale": 7.5,
        "num_inference_steps": 50
      },
      "brand_alignment": {
        "primary_color": "#HEX",
        "mood": "keyword",
        "style": "description"
      }
    },
    {
      "asset_id": "model-001",
      "type": "3d-model",
      "service": "hunyuan3d",
      "prompt": "short english description under 200 chars",
      "parameters": {
        "bbox_condition": [1.0, 0.8, 0.3]
      },
      "brand_alignment": {
        "materials": ["material1"],
        "finish": "matte/glossy"
      }
    }
  ]
}

Also generate a human-readable summary at brands/{BRAND_NAME}/visual-generation-guide.md.

Integration

• Upstream: visual-identity-core, detailed-product-description, product-positioning-summary
• Downstream: visual-asset-generator (consumes the prompts JSON)
• Part of: brand-visual-pipeline orchestrator

CLI Usage

bun scripts/cli.ts activate visual-asset-prompt-generator