Brand Configuration

This guide covers setting up brand data for the brand evaluation gate in the adaptation pipeline.

Brand Model

A Brand record captures the identity, voice, and guidelines that the brand evaluation node checks adapted scripts against.

Field	Description
code	Short identifier, unique (e.g., LAYS, WALKERS, PEPSI)
name	Display name (e.g., “Lay’s”, “Walkers”)
description	Brand overview and positioning statement
guidelines	Free-text brand voice, values, visual identity, and messaging guidelines
insights	Structured brand patterns as [{heading, points[]}] JSON
is_active	Whether the brand is available for selection

Creating a Brand

1. Navigate to TV Spots > Brands > Add Brand

2. Enter a Code and Name

3. Write the Description — positioning and brand essence

4. Write Guidelines — detailed instructions for the brand evaluation gate:

   Voice: Playful, energetic, youthful. Never corporate or stiff.
   Values: Fun, togetherness, everyday moments of joy.
   Visual Identity: Bold primary colors (red, yellow). Product always
   visible. No dark or moody imagery.
   Messaging: Keep taglines simple and memorable. Always end with
   the brand tagline "Life's Good with Lay's."
   Naming: Use market-specific brand name (Walkers in UK, Lay's elsewhere).
   

5. Add Insights for structured brand patterns:

   [
     {
       "heading": "Brand Voice",
       "points": [
         "Playful and conversational tone",
         "Avoid formal or corporate language",
         "Use humor where culturally appropriate"
       ]
     },
     {
       "heading": "Visual Consistency",
       "points": [
         "Product must appear in at least 2 shots",
         "Brand colors (red, yellow) should dominate the palette",
         "Logo visible in opening and closing frames"
       ]
     }
   ]
   

6. Save

How Brand Evaluation Works

The brand evaluation gate is the final evaluation node in the adaptation pipeline. It checks the adapted script against the brand’s guidelines and insights:

1. The pipeline resolves the effective brand for the ad unit (ad unit override → campaign default)

2. The eval-brand prompt template receives the brand’s guidelines and insights fields

3. The LLM evaluates whether the adapted script:

   • Maintains the brand’s voice and tone

   • Preserves visual identity requirements

   • Uses correct brand naming for the target market

   • Aligns with brand values and messaging guidelines

4. If the evaluation fails, the script is sent back to the writer node for revision (up to 3 retries)

See Adaptation Pipeline for the full evaluation gate architecture.

Linking Brands

Brands connect to the adaptation workflow at two levels:

Campaign Level

Every Campaign has an optional Brand FK. This is the default brand for all ad units under the campaign:

1. Navigate to TV Spots > Campaigns

2. Select or create a campaign

3. Set the Brand field

Ad Unit Level (Override)

Individual ad units can override the campaign’s brand. This handles market-specific trade names:

Campaign brand: Lay's (LAYS)

US adaptation: uses campaign brand (Lay's)
UK adaptation: overrides with Walkers (WALKERS)
India adaptation: overrides with Lay's India (LAYS_IN)

The effective_brand property resolves this: ad unit brand first, then campaign brand.

To set an override:

1. Navigate to the adaptation Video Ad Unit

2. Set the Brand field to the market-specific brand

3. Save — the pipeline will use the override brand for evaluation

Guidelines vs. Insights

Both guidelines and insights inform the brand evaluation gate, but serve different purposes:

Field	Format	Best For
guidelines	Free-text	Detailed, nuanced instructions — voice descriptions, naming rules, visual identity requirements, tone guidance
insights	Structured JSON [{heading, points[]}]	Bullet-point patterns — checkable rules, quantifiable requirements, categorical distinctions

Both fields are passed to the eval-brand prompt template. Use guidelines for prose-style brand books and insights for structured evaluation criteria.

Import & Export

Brand data is managed via dedicated commands:

# Export brands to data/brands.json
uv run manage.py export_brands

# Export to a custom directory
uv run manage.py export_brands --dir custom/

# Import brands from data/brands.json
uv run manage.py import_brands

# Preview without importing
uv run manage.py import_brands --dry-run

The JSON format includes all brand fields, referenced by code:

[
  {
    "code": "LAYS",
    "name": "Lay's",
    "description": "Global snack brand...",
    "guidelines": "Voice: Playful, energetic...",
    "insights": [
      {
        "heading": "Brand Voice",
        "points": ["Playful tone", "Avoid formality"]
      }
    ],
    "is_active": true
  }
]