Content Type Router

Content Type Router

Classify a content generation request into a structured content type configuration that provides layout rules, visual mode constraints, typography rules, and generation hints for downstream image generation.

Overview

The router maps natural language descriptions (and optional industry/type hints) to one of 17 predefined content type configurations across 4 categories:

Visual Modes

Every content type has one of three visual modes that determine text handling:

Content Type Reference

Product-Focused (pure_visual)

hero_shot — Single product on clean background, hero positioning

• Focal point: center | Coverage: 60-80% | Background: white/gradient
• Hints: studio lighting, no props, Amazon/Shopify listing standard

lifestyle_shot — Product in real-world context showing aspiration

• Focal point: rule-of-thirds | Coverage: 30-50% | Background: contextual
• Hints: natural lighting, authentic mood, story-driven

flat_lay — Top-down arrangement of multiple items

• Focal point: center | Coverage: 65-75% | Camera: overhead (90°)
• Hints: 3-12 items, cohesive palette, intentional negative space

detail_closeup — Macro shot for texture and craftsmanship

• Focal point: center | Coverage: 80-95% | Camera: macro
• Hints: shallow depth of field, premium material visibility

scale_shot — Product with size reference object

• Focal point: center | Coverage: 40-70% | Reference: hand or common object

unboxing — Product with packaging, reveal aesthetic

• Focal point: center | Coverage: 50-80% | Mood: anticipation/premium

Typography-Heavy (cta)

poster — Designed travel/event poster with headline and imagery

• Text zone: overlay | Fonts: headline + body | Max 7-word headline

sale_promo — Promotional graphic with discount/offer messaging

• Text zone: prominent | Single CTA required | Max 35% text area

countdown — Time-limited offer with countdown element

• Text zone: center | Urgency language | Date/time element required

quote_graphic — Brand quote or testimonial as designed asset

• Text zone: center | Single quote | Minimal supporting context

announcement — Product launch or news announcement graphic

• Text zone: structured | Hierarchy: news > detail > CTA

Educational (educational)

infographic — Data visualization with labeled information

• Text: labels and data points | Max ~35% text area | Clear visual hierarchy

comparison_chart — Side-by-side comparison of options

• Layout: split | Labels required | Clear differentiation

process_howto — Step-by-step visual guide

• Layout: sequential | Step numbers | 3-7 steps recommended

carousel_educational — Multi-frame educational series (single frame rules)

• Consistent style across frames | Progressive information reveal

Social Proof (cta or educational)

testimonial_graphic — Customer quote with attribution

• Quote text required | Customer name/handle | Optional product image

before_after — Side-by-side transformation visual

• Split composition | Clear before/after labeling | High contrast difference

ugc_repost — User-generated content styled for brand repost

• Authentic feel | Minimal brand overlay | Source attribution

Detection API

Python Usage

from content_types.registry import (
    detect_content_type,
    get_content_type_config,
    get_industry_content_types,
    get_content_types_by_visual_mode,
    get_content_types_by_category,
    get_all_content_types,
)
from content_types.base import VisualMode

# Detect from description
slug, confidence = detect_content_type(
    description="I need a product shot on white background for Amazon",
    industry="ecommerce",
)
# → ("hero_shot", 0.95)

# Get full config with layout rules and generation hints
config = get_content_type_config(slug)
prompt_context = config.to_prompt_context()  # Inject into LLM prompt
negative_prompts = config.get_negative_prompt_string()  # For image gen

# Industry-based recommendations
types = get_industry_content_types("beauty")
# → [HeroShot, BeforeAfter, FlatLay]

# Filter by visual mode
pure_visual_types = get_content_types_by_visual_mode(VisualMode.PURE_VISUAL)

# Filter by category
product_types = get_content_types_by_category("product_focused")

Detection Logic

1. If output_type exactly matches a slug → confidence 1.0
2. Keyword scoring across all registered types
3. Normalized score with multi-match bonus
4. If confidence < 0.3 → fall back to industry default
5. Ultimate fallback: lifestyle_shot

Detection Return

(slug: str, confidence: float)
# confidence: 0.0-1.0
# 0.3+ = usable, 0.7+ = high confidence

ContentTypeConfig Structure

@dataclass
class ContentTypeConfig:
    name: str                          # Human-readable name
    slug: str                          # Machine key: "hero_shot"
    category: str                      # "product_focused" | "typography_heavy" | etc.
    definition: str                    # One-line definition for LLM context
    visual_mode: VisualMode            # PURE_VISUAL | EDUCATIONAL | CTA
    layout: LayoutRules                # Focal point, coverage, background, camera angle
    typography: Optional[TypographyRules]  # Headline words, weight, CTA rules
    generation_hints: List[str]        # Positive requirements for image gen prompt
    negative_prompts: List[str]        # What to avoid (comma-joined for diffusion)
    detection_keywords: List[str]      # Keyword matching corpus
    common_industries: List[str]       # Industry affinity
    aspect_ratios: List[str]           # Recommended ratios: ["4:5", "1:1"]

LayoutRules Structure

@dataclass
class LayoutRules:
    focal_point: str           # "center" | "upper_third" | "rule_of_thirds" | "left" | "right"
    text_zone: Optional[str]   # "bottom_40_percent" | "top_20_percent" | "overlay" | None
    subject_coverage_min: float  # Min % of frame for main subject
    subject_coverage_max: float  # Max % of frame for main subject
    text_area_max: float       # Max % of frame for text (0.0 = no text)
    logo_zone: str             # "corner" | "bottom_center" | "none"
    whitespace_min: float      # Minimum negative space target
    camera_angle: Optional[str]  # "eye_level" | "overhead" | "low_angle" | "macro"
    background: Optional[str]  # "white" | "gradient" | "contextual" | "lifestyle"

Industry Defaults

Database Mode

The registry supports optional DB-backed content types (for per-brand customization):

export USE_DB_CONTENT_TYPES=true  # Enable DB mode

When enabled, types load from a content_types Supabase table with brand-specific overrides. Falls back to code-defined types on DB errors.

# Invalidate cache after DB updates
from content_types.registry import invalidate_cache
invalidate_cache()

# Enable/disable at runtime
from content_types.registry import enable_db_mode
enable_db_mode(True)

Integration Pattern

Inject the config into image generation prompts:

slug, confidence = detect_content_type(description, industry=industry)
config = get_content_type_config(slug)

# Build generation prompt
system_context = config.to_prompt_context()
negative = config.get_negative_prompt_string()

# Pass to image generation
generate_image(
    prompt=f"{system_context}\n\n{user_prompt}",
    negative_prompt=negative,
    aspect_ratio=config.aspect_ratios[0],
)