image-generation

name	image-generation
description	Production-grade AI ad imagery — prompt patterns, brand compositing, human-centered psychology, anti-slop standards.
canon_references	["sutherland-signalling","dunford-positioning","godin-permission"]
canon_source	brain/MARKETING_CANON.md
universal	true
note	Examples in this skill may reference SunBiz (legacy client); the skill itself is brand-agnostic. Per-brand context lives in brain/clients/<brand>.md.
triggers	["generate an ad image","create campaign creative","make a social media graphic","AI image for Meta ad","generate ad visuals"]

SKILL: AI Image Generation (Production-Grade)

Generate agency-quality, psychology-based ad creatives with AI-generated humans, logo compositing, and conversion science for SunBiz Funding MCA campaigns.

---

Overview

Production-grade ad creative generation using Google Gemini's native image generation + Pillow logo compositing. Every ad includes photorealistic AI-generated humans for trust, SunBiz Funding logo overlay, conversion psychology principles, and MCA compliance guardrails.

Setup

pip install google-genai Pillow python-dotenv

Add to .env.agents:

GEMINI_API_KEY=your_api_key_here

Place logo at: media/brand/sunbiz_logo.png

Models

Model	Capability	Use
gemini-2.5-flash-image	Nano Banana — fast, production-ready	Primary
gemini-2.0-flash-exp	Legacy experimental	Fallback
imagen-3.0-generate-002	Standalone Imagen API	Final fallback

SDK: google-genai (NOT the deprecated google-generativeai)

Architecture

Master Prompt System

Every prompt is assembled from 6 layers via _build_master_prompt():

1. Scene Prompt — The specific creative direction (human, setting, composition)
2. Logo Instruction — SunBiz logo description + placement rules
3. Color Palette — Brand colors with hex codes
4. Visual Psychology — Z-pattern hierarchy, rule of thirds, contrast, negative space
5. Render Specs — 4K quality, studio lighting, depth of field, typography
6. Compliance — No "loan" language, no guaranteed approvals, disclaimer text

Logo Compositing Pipeline

After Gemini generates the image, Pillow overlays the actual SunBiz logo:

Gemini generates image → save to disk → Pillow opens image + logo →
scale logo to 13% of image width → paste at top-right with alpha →
save final composited image

This ensures pixel-perfect logo placement regardless of what Gemini renders.

Creative Styles

Consolidation Ads (5 styles)

Style	Description	Psychology Trigger
hero_human	Confident business owner with data overlay	Trust + Social Proof
split_screen	Before/after dramatic transformation	Loss Aversion + Relief
data_dashboard	Bloomberg-style financial data viz	Authority + Credibility
testimonial_style	Editorial portrait with quote	Social Proof + Relatability
urgency_relief	Emotional crossroads metaphor	Urgency + Transformation

Growth Capital Ads (3 styles)

Style	Description	Psychology Trigger
ceo_portrait	Executive in premium workspace	Aspiration + Authority
growth_chart	3D growth curve breaking through ceiling	Momentum + Ambition
lifestyle_success	Owner thriving at their business	Aspirational Social Proof

Stories/Reels Vertical (2 styles)

Style	Description	Psychology Trigger
transformation	Cinematic before/after vertical split	Dramatic Transformation
quick_stat	Bold stat with human face close-up	Stop-the-Scroll + Shock Value

Usage

Single Ad

from scripts.imagen_generate import generate_consolidation_ad

result = generate_consolidation_ad(
    headline="STOP THE DAILY DRAIN",
    before_daily="$2,100/day across 4 funders",
    after_daily="$850/day — one partner",
    savings="Save $1,250/day",
    cta="See If You Qualify",
    size="1080x1080",
    campaign_name="consolidation_q2",
    style="hero_human",
)

Growth Capital Ad

from scripts.imagen_generate import generate_growth_ad

result = generate_growth_ad(
    headline="SMART CAPITAL FOR GROWING BUSINESSES",
    cta="See Your Options",
    size="1080x1080",
    campaign_name="growth_q2",
    style="ceo_portrait",
)

Stories/Reels Vertical

from scripts.imagen_generate import generate_stories_ad

result = generate_stories_ad(
    headline="DROWNING IN MCA PAYMENTS?",
    style="transformation",
    campaign_name="stories_q2",
)

Full Campaign Suite (All Styles)

from scripts.imagen_generate import generate_campaign_suite

# Generates ALL consolidation styles in feed + portrait sizes, plus Stories
results = generate_campaign_suite("q2_consolidation", "consolidation")

# Generates ALL growth styles in feed + portrait sizes
results = generate_campaign_suite("q2_growth", "growth")

A/B Test Variants

from scripts.imagen_generate import generate_ad_variants

results = generate_ad_variants(
    base_prompt="Your detailed MCA ad prompt...",
    campaign_name="consolidation_test",
    variants=3,  # Generates 3 conceptually distinct variants
    size="1080x1080",
)

Visual Psychology System

Color Psychology

Color	Hex	Psychological Effect	Use In Ads
Navy Blue	#001F54	Trust, stability, authority	Primary backgrounds
Burnt Orange	#FF6B35	Action, urgency, warmth	CTA buttons
Emerald Green	#28A745	Growth, money, success	Savings/approval indicators
Gold	#D4A843	Premium, value, prosperity	Savings numbers, accents
White	#FFFFFF	Clarity, transparency	Text on dark backgrounds

Human Psychology Triggers

• Hero Human Presence: Increases trust 35% in financial ads — always include people
• Direct Eye Contact: Builds connection and authority with the viewer
• Before/After: Loss aversion (losing $1,250/day) more powerful than gain framing
• Specific Numbers: "$1,250/day saved" converts 3x vs "save money"
• Authority Positioning: Low camera angle, confident posture, professional setting
• Social Proof: Testimonial format with real-looking business owners
• Urgency: Speed messaging ("24-48 hours") with genuine not false scarcity

Visual Composition Rules

• Z-Pattern: Eye flow top-left → headline → center visual → bottom CTA
• Rule of Thirds: Key focal point at grid intersections
• 20% Negative Space: Professional breathing room, not cluttered
• F/2.8 Depth of Field: Sharp subject, soft background = premium feel
• Cinematic Color Grade: Lifted blacks, warm mids, teal shadows

Prompt Engineering Best Practices

DO:

• Describe the scene cinematically (lighting, mood, atmosphere)
• Specify exact camera angles and depth of field
• Include specific color hex codes
• Describe human subjects in detail (age, expression, attire, posture)
• Specify text placement, typography style, and hierarchy
• Include lighting direction and quality
• Reference real-world quality benchmarks ("Think American Express ad quality")

DON'T:

• Use vague prompts ("make a nice ad")
• Skip the human element
• Forget brand colors or logo placement
• Use "loan" language anywhere
• Promise guaranteed approval
• Use red as primary color (signals danger)
• Generate cluttered compositions (>30% text coverage)

Quality Checklist

• Human subject looks realistic and professional (no AI artifacts)
• SunBiz Funding logo visible and properly placed
• Brand colors match (Navy #001F54, Orange #FF6B35, Green #28A745)
• All text readable at mobile size (400px width test)
• Before/after numbers realistic and specific
• MCA disclaimer included
• No "loan" language anywhere
• CTA button prominent and orange
• Visual hierarchy follows Z-pattern
• Image meets platform size requirements
• Emotional tone matches target audience (stress relief for consolidation, ambition for growth)

Platform Sizes

Platform	Size	Ratio	Use
Meta Feed	1080x1080	1:1	Primary placement
Meta Feed	1080x1350	4:5	Optimal feed real estate
Meta Stories/Reels	1080x1920	9:16	Stories and Reels
Google Display	1200x628	1.91:1	Display network

Iteration Process

1. Generate all styles for target audience (5 consolidation or 3 growth)
2. Review — pick top 3 performers by visual quality
3. Regenerate winners with prompt refinements
4. Generate all platform sizes from winning prompts
5. Composite logos via Pillow (automatic)
6. Upload to platforms via media-manager agent
7. Launch as A/B test — after 500+ impressions, measure CPQL
8. Scale winners, iterate losers, maintain 10-15 distinct creatives per campaign