---
name: ad-blueprint-gpt-image-2
description: Generate ultra-detailed ad creative blueprints in JSON format for ANY brand, with paste-ready prompts for GPT Image 2 (ChatGPT Images 2.0). Use this skill (instead of ad-creative-blueprint-generator) any time the user provides a website URL and asks for ad creative production, JSON ad blueprints, ChatGPT/GPT Image prompts, or wants to produce static ads for any product or brand. Triggers on phrases like "fais-moi des JSONs pour [URL]", "produis des concepts pour ce site", "lance la prod créa pour [URL]", "génère des concepts pour [URL]", "/creas", or English equivalents like "make ad blueprints for [URL]".
---

# 🔥 AD CREATIVE BLUEPRINT GENERATOR — Universal Ad Skill

> This skill generates ultra-detailed GPT Image 2 ad creative blueprints in JSON format for **any brand or product**.
>
> **MANDATORY FIRST STEP**: Ask for the website URL → research the brand → confirm the brand profile → then produce JSONs.

---

# TABLE OF CONTENTS

1. Brand Research Protocol (START HERE)
2. Brand Profile Template
3. Avatar Framework
4. Beliefs & Awareness Levels
5. Headlines Framework (H1-H16 universal patterns)
6. Visual Formats (21 formats)
7. Hard Rules (17 rules)
8. Product JSON Template
9. GPT Image 2 Best Practices
10. Press Logos & Amazon Badges
11. Typography & Colors
12. Concept → JSON Workflow
13. Scroll-Stop Psychology
14. Objection Handling
15. Copy Patterns & Microcopy
16. Quality Control
17. JSON Template

---


=================================================================
# SECTION 0 — BRAND RESEARCH PROTOCOL (START HERE)
=================================================================

# MANDATORY FIRST STEP — BRAND RESEARCH

> **Before producing a single JSON, you MUST complete this protocol.**
> Never invent brand data. Every field must come from the website.

---

## HOW TO START

1. **If the user has NOT provided a website URL** → ask:
   *"Donne-moi l'URL du site de la marque et je fais toute l'étude moi-même avant de lancer la prod."*

2. **Once you have the URL** → use WebFetch to scrape the following pages in order:
   - Homepage (`/`)
   - Product page (main product or best-seller)
   - About page (`/about` or `/notre-histoire`)
   - FAQ page (`/faq`) if accessible
   - Reviews section (on product page or dedicated page)

3. **Extract and fill** the Brand Profile Template (Section 1)

4. **Confirm with the user** before generating JSONs:
   *"Voici ce que j'ai extrait du site — confirme ou corrige avant que je lance la prod."*

5. **Only after confirmation** → proceed to Section 12 (Concept → JSON Workflow)

---

## WHAT TO EXTRACT FROM THE WEBSITE

| Field | Where to find it |
|-------|-----------------|
| Brand name | Logo, domain name, meta title |
| Product name & description | Product page hero section |
| Price | Product page (exact price — never approximate) |
| Offer type (BOGO, bundle, discount %) | Product page, homepage banner, cart |
| Guarantee (days + conditions) | Product page, footer, FAQ |
| Shipping offer | Cart/FAQ/footer |
| Core USP / hero claim | Homepage hero headline, product page headline |
| Key differentiators vs competitors | About page, product page comparison sections |
| Target audience signals | Copy tone, imagery, testimonials, model choices |
| Customer testimonials & verbatims | Dedicated reviews section, homepage |
| Competitor mentions | "vs" comparisons, "original" claims, FAQ |
| Press logos / media mentions | Homepage, footer, press page |
| Origin story / legitimacy | About page, product page story section |
| Mechanism / how it works | Product description, dedicated "science" section |
| Social proof numbers | "X+ customers", star ratings, review count |
| Brand voice & tone | Read 3-5 pages of copy |
| Visual aesthetic | Color palette, imagery, design language |

---


=================================================================
# SECTION 1 — BRAND PROFILE TEMPLATE
=================================================================

> Fill every field after website research. This becomes the single source of truth injected in all JSONs.
> Never leave a field blank — if not found on site, derive from context or mark as [NOT FOUND — derive].

## Company

- **Brand name**: [extracted from site]
- **Domain**: [URL provided by user]
- **Legal / ownership**: [if visible on site]
- **E-commerce platform**: [Shopify / WooCommerce / etc. — detectable via page source]

## Product

- **Product name**: [extracted]
- **One-line description**: [extracted from product page hero]
- **Price**: **$[X]** (FIXED — always use this EXACT price in all JSONs and ad prompts — never round or approximate)
- **Offer**: [e.g. Buy One Get One Free / Bundle 3 for $X / 50% OFF]
- **Guarantee**: [e.g. 30-day money-back guarantee — extract exact conditions]
- **Shipping**: [e.g. Free shipping — extract exact conditions]
- **Key material / ingredient / technology**: [what makes it work — extracted]
- **Magnetic or mechanism claim**: [extracted from product description]

## Positioning

### What the product IS
- [3-5 positioning statements extracted from site copy]

### What the product is NOT
- [3-5 anti-positioning statements — look for "not a…", "unlike…", competitor comparisons]

## USP (Unique Selling Proposition)

> [Single clearest benefit statement — often the homepage hero headline]

## Mechanism story

**[Brand's proprietary mechanism name if any]**

[Extract the "how/why it works" explanation from the site. This is the rational justification for purchase.]

## Tagline options

- [Extracted from site, or derived from copy — 3-5 options]

## Pricing structure

| Offer | Price | Crossed price | Savings |
|-------|-------|---------------|---------|
| [extract all tiers] | | | |

**Hero offer**: [Best-value / most-promoted offer]

## Voice & tone

- [Derive from reading 3-5 pages: formal/casual, masculine/feminine, premium/accessible, aggressive/soft, scientific/emotional]

## Social proof numbers

- [X]+ customers
- [X]-star rating from [Y] reviews
- [Other proof numbers from site]


=================================================================
# SECTION 2 — AVATAR FRAMEWORK
=================================================================

> Build the target customer avatar using this framework after website research.
> Clues: testimonials, copy tone, imagery, product name, price point, FAQ questions, model choices.

## Identity

- **Name**: [assign a persona name — relatable for the target audience]
- **Internal nickname**: [what they want — e.g. "The man who wants his edge back" / "The woman who wants her confidence back"]
- **Defining phrase**: [their internal monologue — derive from testimonials and copy tone]

## Demographics

| Field | Value |
|-------|-------|
| Age | [derive from price point + copy + imagery] |
| Gender | [derive from imagery + copy] |
| Family situation | [derive from copy themes] |
| Location | [derive from site language + shipping countries] |
| Occupation | [derive from copy + price point] |
| Income | [derive from price point + aesthetic level] |
| Education | [derive from vocabulary level in copy] |

## Psychographics

### Personality & values
- [derive from copy tone — 3-5 traits]

### Interests
- [derive from influencer mentions, media references in copy]

### Media consumed
- [derive from press logos, influencer references, platform choices]

## Current state — "BEFORE" (pains & frustrations)

### Daily frustrations
- [extract from product page pain points — 5-7 items]

### What keeps them up at night
- [derive from deep fear language in copy — 3-5 items]

### What they've already tried (without success)
- [extract from "you've tried X but…" type copy — 3-5 items]

## Desired state — "AFTER" (aspirations & dreams)

### How they want to feel
- [extract from benefit copy — 5 items]

### Concrete results they hope for
- [extract from testimonials and outcome claims — 5 items]

## Deep fears

- [derive from pain-point copy — 4-5 items]

## Objections to purchase

| Objection | Counter-message |
|-----------|-----------------|
| "It's a placebo / doesn't really work" | [extract from site's proof mechanism] |
| "Too expensive" | [extract from site's value comparison or ROI argument] |
| "No proof" | [extract from testimonials / social proof / mechanism] |
| "I can find cheaper elsewhere" | [extract from differentiation claims] |
| "How long before I see results?" | [extract from testimonials with specific timelines] |

## Customer language & vocabulary

### Words they use to describe their problems
- [extract from testimonials — 5-7 direct quotes or paraphrases]

### Words they use to describe their desires
- [extract from testimonials and aspirational copy]

## Customer verbatims (extract from testimonials — use these in copy)

- *"[Testimonial 1]"*
- *"[Testimonial 2]"*
- *"[Testimonial 3]"*
- *"[Testimonial 4]"*
- *"[Testimonial 5]"*

## Purchase journey

### How they discover the brand
- [derive from ad platform signals, SEO keywords in meta tags, referral mentions]

### What makes them click
- [derive from hero section copy and key hooks]

### What convinces them to buy
- [derive from proof sections, guarantee, testimonials, offer]

## Strategic summary

[Write a 3-4 sentence avatar summary after research. Who is this person? What do they feel? What solution are they looking for? How does this brand position as the answer?]


=================================================================
# SECTION 3 — BELIEFS AND AWARENESS LEVELS
=================================================================

# THE 6 CORE BELIEFS + 5 LEVELS OF AWARENESS

## Marketing principle (foundation)

> Marketing = transforming the prospect's current beliefs into beliefs that align with them and naturally lead them to buy.
>
> Each ad/section installs ONE belief in sequence. The skeptical prospect is guided belief by belief until purchase becomes the only logical conclusion.
>
> *"Stop writing copy. Start building arguments."*

---

## THE 6 CORE BELIEFS (Universal Beliefs Ladder)

> These 6 beliefs are universal for any product. After website research, fill in the brand-specific copy using extracted language.

### Belief #1 — THE PROBLEM IS SOLVABLE
> *"My [main pain point from site] is not a fatality — it's a situation that can be corrected."*

- **Why crucial**: If the prospect believes nothing can be done, they'll never buy. They must first believe a solution exists.
- **Function**: Opens the door to a solution
- **Page section**: Hero / Headline
- **Copy hooks**: [derive from site — problem-reframe angles, "it's not X, it's Y" structures]

### Belief #2 — CLASSIC SOLUTIONS FAIL
> *"[Competing solutions from site] are not the right solution — they [fail in specific ways]."*

- **Why crucial**: The prospect has already tried alternatives. If they still believe those work, they'll go back to them instead of trying this.
- **Function**: Closes competing alternatives
- **Page section**: Problem / Agitation
- **Copy hooks**: [derive from site — anti-competitor language, "stop doing X" angles]

### Belief #3 — LEGITIMACY / ORIGIN
> *"[Brand's origin story, credentials, or scientific basis] is why this solution has worked for [X] people."*

- **Why crucial**: Bypasses skepticism by leaning on historical, scientific, or social authority.
- **Function**: Creates legitimacy and credibility
- **Page section**: Brand story / How it works / Social proof
- **Copy hooks**: [derive from site — origin story, certifications, reviews, media mentions]

### Belief #4 — THE MECHANISM WORKS
> *"The [product mechanism extracted from site] creates a real effect — I can understand why it works."*

- **Why crucial**: Without a credible mechanism, the prospect can't rationally justify their purchase.
- **Function**: Justifies the purchase rationally
- **Page section**: How it works / Science / Mechanism
- **Copy hooks**: [derive from site — mechanism explanation, specific technical claims]

### Belief #5 — DIFFERENTIATION
> *"[Brand name] is the original/best quality — and cheaper or competing alternatives don't work because [reason from site]."*

- **Why crucial**: If the prospect believes all options in the category are equal, they'll buy the cheapest. They must believe only this brand delivers.
- **Function**: Eliminates direct competition
- **Page section**: Differentiation / "We Are The Original"
- **Copy hooks**: [derive from site — comparison language, quality claims, "original" positioning]

### Belief #6 — RISK-FREE TRIAL
> *"I can try [brand name] risk-free — if it doesn't work for me, I get a full refund."*

- **Why crucial**: Neutralizes the last barrier. Transforms the purchase into a no-consequence test.
- **Function**: Removes the last objection
- **Page section**: Guarantee / Risk reversal
- **Copy hooks**: [from site — exact guarantee language]

---

## RECAP TABLE

| # | Belief | Function in argument |
|---|--------|----------------------|
| 1 | The problem is solvable | Opens the door to a solution |
| 2 | Classic solutions fail | Closes alternatives |
| 3 | Legitimacy / origin | Creates credibility |
| 4 | The mechanism works | Justifies rationally |
| 5 | Brand is different | Eliminates competition |
| 6 | No risk | Removes the last objection |

## INSTALLATION SEQUENCE ON SALES PAGE

| Page section | Belief installed |
|-------------|------------------|
| 1. Hero / Headline | Belief #1 — The problem is solvable |
| 2. Problem / Agitation | Belief #2 — Classic solutions fail |
| 3. Brand story / Social proof | Belief #3 — Legitimacy |
| 4. How it works / Mechanism | Belief #4 — The mechanism works |
| 5. Differentiation | Belief #5 — Brand is different |
| 6. Guarantee / Risk reversal | Belief #6 — No risk |
| 7. CTA / Final offer | ALL beliefs converge → PURCHASE |

**EVERY JSON ad blueprint must map to one or more of these beliefs in the `belief_installation` field.**

---

## THE 5 LEVELS OF AWARENESS (Schwartz framework)

| Level | State | Strategy | Headlines that fit |
|-------|-------|----------|---------------------|
| **L1 — Unaware** | Doesn't know they have a problem | Make them recognize the problem via a specific, painful moment | Situational/future angles |
| **L2 — Problem Aware** | Knows the problem, doesn't know the cause | Name the cause (the real underlying mechanism) | H6, H10, H11, H12, H13, H16 patterns |
| **L3 — Solution Aware** | Knows solutions exist, comparing categories | Position brand vs competing solution categories | H5, H8, H14, H15 patterns |
| **L4 — Product Aware** | Knows the product category, doubts quality | Differentiate (Original vs Copies, quality proof, specs) | H4, H9 patterns |
| **L5 — Most Aware** | Ready to buy, needs offer/urgency | Offer + countdown + risk reversal | H1, H2, H3 patterns |

**EVERY JSON must specify the consciousness level (1-5) in the `consciousness_level` field.**

## Audience volume by level (general guideline)

- **L1** (Unaware): ~60-70% of cold audience — biggest opportunity, hardest to crack
- **L2** (Problem Aware): ~20% — sweet spot for scaled campaigns
- **L3** (Solution Aware): ~7% — high-intent, comparing alternatives
- **L4** (Product Aware): ~2% — already considering this product category
- **L5** (Most Aware): ~1% — retargeting / email / abandoned cart

**Strategic implication**: Ad creative must cover ALL 5 levels to capture the full funnel.


=================================================================
# SECTION 4 — HEADLINES FRAMEWORK
=================================================================

# THE H1-H16 UNIVERSAL HEADLINE PATTERN LIBRARY

> **HOW TO USE**: These are universal headline PATTERNS. After website research (Section 0), fill in the `[BRAND VARIABLE]` placeholders with brand-specific content extracted from the site.
>
> **CRITICAL RULE**: This is the SOURCE OF TRUTH for every ad. Never invent a headline from scratch. Always pull from H1-H16 patterns (or their pre-approved variations). Preserve the hook structure — adapt the words to the brand.

---

## CATEGORY A — OFFER / URGENCY

### **H1 — [OFFER HOOK]**
- **Pattern**: `[CORE OFFER IN DIRECT TERMS] — [URGENCY TRIGGER]`
- **Example fill**: "BUY 1 + GET 1 FREE — BOGO ENDS TONIGHT"
- **Sub**: "[Offer deadline or scarcity signal]"
- **Best for**: floating product shots, gradient backgrounds, product scattered around hero element
- **Beliefs installed**: #6
- **Awareness level**: L5
- **Angle**: Offer Urgency / Last Chance
- **Fill from site**: Use the brand's exact offer language (BOGO, bundle, % off, free gift)
- **Variations** (adapt to brand):
  - GRAB 2 — PAY FOR 1
  - DOUBLE THE [BENEFIT] — HALF THE PRICE
  - 2 [PRODUCTS]. 1 PRICE. TODAY ONLY.
  - YOUR SECOND [PRODUCT] IS ON US
  - [OFFER NAME] LIVE — STOCK ENDS AT MIDNIGHT
  - PAY ONCE. [RESULT] TWICE.
  - ONE FOR YOU. ONE ON THE HOUSE.
  - TWO [PRODUCTS]. ZERO EXTRA COST.

### **H2 — LAST CHANCE — SALE ENDS IN [X] HOURS!**
- **Pattern**: `LAST CHANCE — SALE ENDS IN [X] HOURS!`
- **Sub**: "[Key benefit 1] / [Key benefit 2] / [Results Guarantee]"
- **Best for**: bright saturated background, product in use or premium packaging
- **Beliefs installed**: #4, #6
- **Awareness level**: L5
- **Angle**: Urgency + Core Benefit Stack
- **Fill from site**: Replace sub-headline benefits with the brand's 2-3 core claims
- **Variations**:
  - [X] HOURS LEFT — THEN IT'S GONE
  - CLOSING IN [X] MINUTES
  - FINAL CALL — DEAL DIES TONIGHT
  - DON'T BLINK — SALE ENDS AT [TIME]
  - THE CLOCK IS RUNNING OUT
  - MIDNIGHT DEADLINE — NO EXTENSIONS
  - LAST [X] HOURS TO LOCK IN [OFFER]
  - AFTER TONIGHT — FULL PRICE RETURNS

### **H3 — SAVE $[AMOUNT] ON YOUR [TRANSFORMATION]**
- **Pattern**: `SAVE $[PRICE] ON YOUR [CORE TRANSFORMATION]`
- **Sub**: "[Key transformation] in [timeframe]. [Offer name]."
- **Best for**: mini web-page format on clean background with CTA button
- **Beliefs installed**: #6
- **Awareness level**: L5
- **Angle**: BOGO/Discount + Transformation Proof
- **Fill from site**: Use the exact discount amount and the brand's core transformation claim
- **Variations**:
  - POCKET $[X] — KEEP THE [TRANSFORMATION]
  - YOUR $[X] DISCOUNT IS WAITING
  - $[X] OFF. ZERO STRINGS.
  - CUT $[X] — UNLOCK EVERYTHING
  - $[X] CLOSER TO [DESIRED OUTCOME]
  - TAKE $[X] OFF — [TRANSFORM] IN [TIMEFRAME]
  - $[X] SAVED. [CORE DREAM RESULT] RECLAIMED.
  - YOUR [TRANSFORMATION] JUST GOT $[X] CHEAPER

---

## CATEGORY B — DIFFERENTIATION / ANTI-COMPETITOR

### **H4 — THE ORIGINAL VS THE COPYCATS**
- **Pattern**: `THE ORIGINAL VS THE COPYCATS`
- **Sub**: "Don't Get Fooled — Get The Original"
- **Best for**: side-by-side comparison + spec table (brand vs inferior alternatives)
- **Beliefs installed**: #5
- **Awareness level**: L4
- **Angle**: Us vs Them / Anti-[Competitor or generic category]
- **Fill from site**: Extract the brand's exact differentiation claims and the specific ways competitors fall short
- **Variations**:
  - THE REAL [BRAND] VS THE FAKES
  - ACCEPT NO COPIES — ACCEPT NO COMPROMISES
  - WHY 9 OUT OF 10 [PRODUCT CATEGORY] [PRODUCTS] ARE [INFERIOR]
  - [BRAND] VS [PLATFORM/COMPETITOR] — THE BRUTAL TRUTH
  - THE ONLY [PRODUCT] WORTH [USING/WEARING/BUYING]
  - STOP BUYING [INFERIOR ALTERNATIVE] — DEMAND THE ORIGINAL
  - THE DIFFERENCE BETWEEN [$CHEAP] AND THE REAL DEAL
  - COPIED [N] TIMES. NEVER MATCHED.

### **H5 — [OLD SOLUTION] IS OVER.**
- **Pattern**: `[COMPETING/OLD SOLUTION] IS OVER.`
- **Sub**: "Or money back guaranteed."
- **Best for**: pure dark background, product centered + 4 corner callouts + press logo strip
- **Beliefs installed**: #2, #4
- **Awareness level**: L3
- **Angle**: Category Killer / Anti-[Old Competing Solution] + Press Authority
- **Fill from site**: Extract what competing solutions or behaviors this product replaces
- **Variations**:
  - SAY GOODBYE TO [OLD SOLUTION]
  - [OLD SOLUTION] IS DEAD
  - WHY [TARGET AUDIENCE] ARE QUITTING [OLD SOLUTION] IN [YEAR]
  - [OLD SOLUTION] RETIRED. [BRAND] HIRED.
  - THE END OF THE [OLD SOLUTION] ERA
  - [OLD SOLUTION] OUT. [CORE BENEFIT] IN.
  - PILLS OUT. POWER IN. *(adapt category as needed)*
  - BURN YOUR [OLD SOLUTION].

---

## CATEGORY C — DESIRE / EMOTIONAL OUTCOME

### **H6 — THEY WON'T SAY IT. BUT THEY'LL FEEL IT.**
- **Pattern**: `[THIRD PARTY] WON'T SAY IT. BUT THEY'LL FEEL IT.`
- **Sub**: "What [product mechanism or key ingredient] does for [avatar]:"
- **Best for**: product against atmospheric background + 5 arrowed benefit bullets + footer offer
- **Beliefs installed**: #4
- **Awareness level**: L2
- **Angle**: Social Proof via Third-Party Reaction / Desire
- **Fill from site**: Adapt third party (partner, boss, friends) to fit the avatar's social context; extract benefit bullets from site
- **Variations**:
  - [THEY]'LL NEVER ASK. BUT [THEY]'LL NOTICE.
  - [THEY] WON'T MENTION IT. [THEY]'LL JUST [REACT DIFFERENTLY].
  - [THEY]'LL NOTICE BEFORE YOU DO.
  - [THEY] WON'T COMPLIMENT IT. [THEY]'LL [CRAVE/FEEL] IT.
  - THE CHANGE [THEY]'LL FEEL BEFORE [THEY] SEE.
  - [THEY] WON'T ADMIT IT. BUT [THEY]'LL WANT MORE.
  - SILENT TO [THEM]. UNFORGETTABLE TO BOTH.

### **H7 — I'M [HAVING THE BEST RESULT] AT [AGE/TIMEFRAME]**
- **Pattern**: `I'M [EXPERIENCING THE BEST {CORE OUTCOME}] AT [AGE OR TIME CONTEXT].`
- **Sub**: "Or get your money back."
- **Best for**: atmospheric gradient background, product centered + 4 callouts (key benefits) + press logo strip
- **Beliefs installed**: #6
- **Awareness level**: L2
- **Angle**: First-Person Confession / Press Authority
- **Fill from site**: Extract the most dramatic outcome claim from testimonials or product copy; adapt age/timeframe to avatar
- **Variations**:
  - NEVER FELT THIS GOOD. NOW I DO AT [AGE].
  - MY [TIME PERIOD] ARE BETTER THAN [PAST COMPARISON].
  - [AGE] AND OUTLASTING MY [COMPARISON].
  - THE BEST [CORE RESULT] OF MY LIFE STARTED AT [AGE].
  - [AGE] AND [UNEXPECTED IMPRESSIVE OUTCOME].

### **H8 — [TIMEFRAME] TO [DESIRED RESULT]**
- **Pattern**: `[SHORT TIMEFRAME] TO [DESIRED RESULT]`
- **Sub**: "Without [Old Solution] Or Side Effects. OR IT'S FREE"
- **Best for**: atmospheric product shot, product centered + 4 callouts (benefits) + press logos
- **Beliefs installed**: #4, #6
- **Awareness level**: L3
- **Angle**: Fast Results / Anti-[Old Solution] / Press Authority
- **Fill from site**: Extract result timeframes from testimonials or product page claims
- **Variations**:
  - [X HOURS/DAYS] TO [RESULT]
  - [X] HOURS — AND IT'S BACK
  - [RESULT] RETURNS IN [TIMEFRAME]
  - [TIMEFRAME]. THAT'S ALL IT TAKES.
  - MISSING [RESULT]? IT'S BACK IN [TIMEFRAME].
  - REBUILD [RESULT] IN [TIMEFRAME]
  - [TIMEFRAME] TO [FEEL/LOOK/BE] LIKE [COMPARISON]

---

## CATEGORY D — TRANSFORMATION / MULTI-BENEFIT

### **H9 — ONE [PRODUCT]. [N] PROBLEMS SOLVED.**
- **Pattern**: `ONE [PRODUCT]. [N] PROBLEMS SOLVED.`
- **Sub**: "[Problem 1] / [Problem 2] / [Problem 3] / [Problem 4] / [Problem 5]"
- **Best for**: dark background, product on one side + N benefit bullets on other side with arrows + footer offer
- **Beliefs installed**: #4
- **Awareness level**: L4
- **Angle**: Multi-Benefit Stack / Problem Elimination
- **Fill from site**: Extract ALL key benefits from product page — list them as the sub-headline problems solved
- **Variations**:
  - ONE [PRODUCT]. [N] FIXES.
  - [N] [AVATAR'S] PROBLEMS — ONE [PRODUCT]
  - THE FIX FOR EVERYTHING [AVATAR] WON'T TALK ABOUT
  - [N] PROBLEMS. [MECHANISM]. ONE SOLUTION.
  - ONE [PRODUCT] REPLACES YOUR ENTIRE [COMPETING STACK]
  - THE [MECHANISM/INGREDIENT] ANSWER TO [N] SILENT BATTLES

### **H10 — Struggling with [MAIN PAIN POINT]?**
- **Pattern**: `Struggling with [MAIN PAIN POINT]?`
- **Sub**: "NO MORE: [Pain 1] / [Pain 2] / [Pain 3]"
- **Best for**: light gradient background, product on one side + 3 pain icons on other with arrows + CTA button at bottom
- **Beliefs installed**: #1
- **Awareness level**: L2
- **Angle**: Pain Trigger / Problem Agitation
- **Fill from site**: Extract the avatar's main pain points from product page and testimonials
- **Variations**:
  - TIRED OF [PAIN]?
  - EMBARRASSED BY [PAIN]?
  - [PAIN] RUINING YOUR [LIFE AREA]?
  - WHY YOU STOPPED [POSITIVE ACTIVITY] BECAUSE OF [PAIN]
  - THE [PAIN] PROBLEM NO ONE ADMITS
  - STILL DEALING WITH [PAIN]?

### **H11 — [SPECIFIC METRIC] OF [PROBLEM] GONE**
- **Pattern**: `[SPECIFIC NUMBER + UNIT] OF [PROBLEM] — [VANISHED/GONE]`
- **Sub**: "[Benefit 1] / [Benefit 2] / [Benefit 3] / [Benefit 4]"
- **Best for**: before/after illustration or metric proof visual + product; footer "Or get your money back"
- **Beliefs installed**: #1, #4
- **Awareness level**: L2
- **Angle**: Specific Number Hook / Transformation / Risk Reversal
- **Fill from site**: Extract specific numbers from testimonials (weight lost, time saved, measurements improved, etc.)
- **Variations**:
  - [X UNIT] OF [PROBLEM] — VANISHED
  - YOUR [PROBLEM] GONE IN [TIMEFRAME]
  - [SPECIFIC METRIC] [LIGHTER / BETTER / FASTER / GONE]
  - [PROBLEM] ERASED — [POSITIVE RESULT] RETURNED
  - [X] [UNIT] [PROBLEM]. [TIMEFRAME]. NO [COMPETING METHOD].

---

## CATEGORY E — PAIN / FUNCTIONAL PROBLEM

### **H12 — [X YEARS/UNITS] OF [PAIN TYPE] GONE IN [TIMEFRAME]**
- **Pattern**: `[X YEARS/UNITS] OF [PAIN TYPE] GONE IN [TIMEFRAME]`
- **Sub**: *(dramatic silence — headline IS the entire promise)*
- **Best for**: scientific/anatomical or clinical visual + product as the solution element
- **Beliefs installed**: #1, #4
- **Awareness level**: L2
- **Angle**: Pain Elimination / Specific Number Hook
- **Fill from site**: Extract pain types addressed and timeframe claims from testimonials
- **Variations**:
  - [X] YEARS OF [PAIN] — ERASED IN [TIMEFRAME]
  - A DECADE OF [PROBLEM]. GONE IN [TIMEFRAME].
  - [X] YEARS OF [PAIN TYPE] — VANISHED IN A [WEEK/MONTH]
  - THE [TIMEFRAME] FIX FOR [X] YEARS OF [PAIN]
  - WALK / SLEEP / LIVE [PAIN-FREE OUTCOME] IN [TIMEFRAME] — OR YOUR MONEY BACK
  - [PAIN TYPE] GONE BEFORE YOUR NEXT [COMPETING APPOINTMENT/REFILL]
  - [TIMEFRAME] TO [PAIN-FREE STATE OUTCOME]

---

## CATEGORY F — HUMOR / MEME / VIRAL

### **H13 — [SPECIFIC CONTEXT]. [UNEXPECTED OUTCOME].**
- **Pattern**: `[TIME OR SPECIFIC SITUATION]. [IRONIC / SURPRISING / ASPIRATIONAL OUTCOME].`
- **Sub**: integrated speech bubbles or captions reacting to the situation
- **Best for**: illustrated cartoon or meme-style creative; humor makes taboo/desire subject acceptable
- **Beliefs installed**: #4, #6
- **Awareness level**: L2
- **Angle**: Humor / Viral Meme / Desire Made Palatable Through Comedy
- **Fill from site**: Adapt to the product's core desire/outcome through a humorous or relatable lens
- **Variations**:
  - [ABSURD TIME/PLACE]. [PRODUCT RESULT].
  - [RELATABLE EMBARRASSING SITUATION]. [UNEXPECTED SOLUTION].
  - [SPECIFIC CLOCK TIME]. [SURPRISING ONGOING OUTCOME].

### **H14 — [THIRD PARTY] FOUND [AVATAR]'S SECRET.**
- **Pattern**: `[THIRD PARTY] FOUND [AVATAR]'S [OUTCOME] SECRET.`
- **Sub**: "It's not a [old solution]. It's not a [old solution 2]. It's a $[PRICE] [product description]."
- **Best for**: illustrated or realistic discovery scene — curious third party holding/finding the product
- **Beliefs installed**: #2, #4
- **Awareness level**: L3
- **Angle**: Anti-[Old Solution] / Curiosity Hook / "Discovery" Narrative Format
- **Fill from site**: Extract the competing alternatives the brand replaces; use exact price from site

### **H15 — [THIRD PARTY] BEGS FOR [MORE / ROUND N]**
- **Pattern**: `[THIRD PARTY] BEGS FOR [ANOTHER / MORE / ROUND N].`
- **Sub**: "No [old solution]. [Core result] in [timeframe]."
- **Best for**: editorial-level realistic photo or premium visual + product overlay + press logos
- **Beliefs installed**: #4, #6
- **Awareness level**: L3
- **Angle**: Core Desire / Press Authority / Aspirational Outcome
- **Fill from site**: Extract desire-outcome and "without X" language from product page
- **Variations**:
  - [THEY]'RE BEGGING FOR [MORE].
  - [THEY] WON'T STOP AT [FIRST ROUND].
  - [N] [ROUNDS/TIMES]. ZERO [OLD SOLUTION].
  - [THEY] ASKED FOR MORE. [AVATAR] DELIVERED.
  - [AVATAR] STOPPED COUNTING. [THEY] DIDN'T.

### **H16 — A [MEASURABLE VISIBLE PROOF] EVERY [TIMEFRAME].** 👇
- **Pattern**: `A [MEASURABLE VISIBLE PROOF] EVERY [TIMEFRAME].`
- **Sub**: "I started [baseline state]. I'm [improved state]."
- **Best for**: object metaphor showing measurable progress (belt holes, scale, before/after proof object, milestone tracking)
- **Beliefs installed**: #1, #4
- **Awareness level**: L2
- **Angle**: Visual Proof / Object Metaphor / First-Person Confession
- **Fill from site**: Extract measurable proof claims from testimonials; find the most tangible, visual proof of progress
- **Variations**:
  - DOWN [X UNITS] IN [TIMEFRAME].
  - FROM [BASELINE] TO [GOAL] — IN [TIMEFRAME].
  - [X UNITS] [BETTER / LIGHTER / FASTER]. NO [COMPETING METHOD].
  - MY [PROOF OBJECT] TELLS THE TRUTH. SO DOES MY [MIRROR / BODY / RESULTS].
  - FROM [NEGATIVE STATE] TO [POSITIVE STATE] — IN [TIMEFRAME].
  - [PROOF METRIC] SMALLER/BETTER. [TIMEFRAME]. NO [GYM / PILLS / SURGERY].

---

## QUICK COMPATIBILITY MATRIX

| Headline | Levels | Beliefs | Best Format Categories |
|----------|--------|---------|------------------------|
| H1 | L5 | #6 | Product Hero, Flat Lay, BOGO |
| H2 | L5 | #4, #6 | Product Shot, Premium Box, Wrist/Use Shot |
| H3 | L5 | #6 | Marble Mini Web Page, Body Transformation |
| H4 | L4 | #5 | Comparison Bar, Spec Table, Anti-Competitor |
| H5 | L3 | #2, #4 | Press Authority, Pure Dark Hero |
| H6 | L2 | #4 | Bullet Layout, Atmospheric Background, Sensual |
| H7 | L2 | #6 | Press Authority, Confession, Gradient |
| H8 | L3 | #4, #6 | Press Authority, Atmospheric, Sensual |
| H9 | L4 | #4 | Bullet Layout, Multi-Benefit Stack |
| H10 | L2 | #1 | Bullet Layout, Pain Icons, Body Diagram |
| H11 | L2 | #1, #4 | Before/After, Transformation, Metric Visual |
| H12 | L2 | #1, #4 | Anatomical/Scientific, Pain Map |
| H13 | L2 | #4, #6 | Cartoon/Meme Bedroom or Situation |
| H14 | L3 | #2, #4 | Cartoon/Meme Discovery, Realistic Discovery |
| H15 | L3 | #4, #6 | Sensual Realism + Press, Bullet + Real Photo |
| H16 | L2 | #1, #4 | Object Metaphor, Confession, Before/After Proof |

## ADAPTATION RULES

When a concept doesn't directly match a headline's `Best for` field but the angle is relevant:

1. **Preserve the hook structure** (specific number, scroll-stop punch, taboo-friendly tone)
2. **Replace the outcome word** to fit the brand's core result
3. **Keep the timeframe specificity** (or tighten it: 72h → 24h, 21 days → 14 days)
4. **Keep the risk reversal** if present
5. **Always pull exact claims from brand research** (Section 1) — never invent

## PROMISE STRENGTHENING (apply on every headline)

- **72h → 24h**, **3 nights → 1 night**, **21 days → 14 days**, **9 days → 7 days** when scroll-stop demands it
- **Vague → Specific**: "better results" → "deepest [outcome] of your life", "more energy" → "[avatar's prime age]-level energy"
- **Keep weird specific numbers**: 43 seconds, 825 grams, 6 years, 9 days (high believability — oddly specific = credible)
- **Always pair aggressive claims with risk reversal** in sub-headline or footer


=================================================================
# SECTION 5 — VISUAL FORMATS
=================================================================

# VISUAL FORMATS LIBRARY — Concept-to-Headline Mapping

> Every ad concept must match one (or a hybrid of two) of these 21 visual formats.
> Each format description uses `[PRODUCT]` as a placeholder — replace with the actual product from site research.

---

## FORMAT 1 — PRODUCT HERO

**Description**: Single [PRODUCT] floating centrally, premium studio lighting, gradient or seamless background. Pure product-focused, no humans.

**Visual signature**:
- [Product] centered, optimal orientation for the product type
- 1080×1080 square, product occupying ~50% of frame
- Diffused softbox lighting catching product surface details
- Soft contact shadow beneath product
- Background: white/gray gradient, pastel, or pure black
- Optional: scattered complementary elements around product

**Compatible headlines**: H1, H2, H3, H9, H12

**Amazon badge**: include

**Human subjects**: none

---

## FORMAT 2 — IN-USE / LIFESTYLE SHOT

**Description**: [PRODUCT] being used by a person in a premium lifestyle context.

**Visual signature**:
- Person using/wearing the product authentically
- Premium wardrobe and setting consistent with brand aesthetic
- Natural light or cinematic studio lighting
- Background: lifestyle context appropriate to brand (office, home, gym, outdoors)
- Product clearly visible and featured

**Compatible headlines**: H2, H6, H10

**Amazon badge**: include

---

## FORMAT 3 — COMPARISON BAR

**Description**: Side-by-side spec table format showing [BRAND] Original vs inferior alternatives/competitors.

**Visual signature**:
- Top: massive headline "THE ORIGINAL VS THE COPYCATS" or equivalent
- Middle: inferior product (left, dull) | "VS" | brand product (right, premium/highlighted)
- Spec table below in rows with comparison data:
  - Key quality spec: [INFERIOR] (red) vs [BRAND] (green)
  - Key performance spec: [INFERIOR] (red) vs [BRAND] (green)
  - Guarantee: NO (red) vs [X] DAYS (green)
  - Verified results: NO (red) vs [X+] (green)
- Bottom: CTA bar "Don't Get Fooled — Get The Original"
- Background: pure black preferred

**Compatible headlines**: H4

**Amazon badge**: exclude (format IS the authority signal)

---

## FORMAT 4 — BULLET LAYOUT

**Description**: [PRODUCT] on one side + arrowed bullet points on the other listing benefits. Most versatile high-converting format.

**Visual signature**:
- Product on left or right, large, dominating ~40% of frame
- 4-5 bullets with arrows pointing from bullets to product or vice versa
- Arrows: thin curved or straight, white or light gray
- Each bullet: bold benefit + relevant icon/emoji
- Optional sub-text under each bullet in lighter weight
- Background: dark (black, deep purple, dark green) or atmospheric
- Footer strip: discount + free shipping + guarantee badges

**Compatible headlines**: H6, H9, H10, H11, H12, H15

**Amazon badge**: include

---

## FORMAT 5 — PRESS AUTHORITY

**Description**: [PRODUCT] centered with press logo strip at bottom. Classic credibility play.

**Visual signature**:
- Product centered (sometimes inside or surrounded by content area)
- 4 corner callouts with thin white arrows pointing to product
- Bottom strip (10-15% of frame): horizontal row of grayscale press logos
- Background: pure black, dark gradient, or moody atmospheric
- Often paired with risk reversal in sub-headline

**Compatible headlines**: H5, H7, H8, H15

**Amazon badge**: exclude (press logos serve the same authority function)

---

## FORMAT 6 — ANATOMICAL / SCIENTIFIC

**Description**: Body silhouette (anatomical, X-ray-style) or scientific visual showing affected areas/mechanism + [PRODUCT] as the solution.

**Visual signature**:
- Neon outlined body silhouette or scientific diagram (on dark background)
- Highlighted problem zones
- Particles/energy effects around the body or diagram
- Product floating beside or below as the solution
- Background: dark with subtle particles or clinical white
- Headline in bold white/red mix
- Futuristic, science-backed mood

**Compatible headlines**: H12 (primary), adapt H6/H8 for mechanism-mapping

**Amazon badge**: exclude (clinical aesthetic)

---

## FORMAT 7 — BEFORE/AFTER

**Description**: Visual transformation shown side-by-side via illustration, photo, or icon.

**Visual signature**:
- BEFORE state left/top: the problem visually represented
- AFTER state right/bottom: the desired outcome
- Product as the bridge between states (floating or referenced)
- Labels: "BEFORE" / "AFTER" or clear visual distinction
- Arrow or divider between states

**Compatible headlines**: H11 (primary), adapt H10, H16

---

## FORMAT 8 — CONFESSION / FIRST-PERSON

**Description**: Large first-person headline dominates. Product smaller/secondary. Minimal layout — the copy IS the ad.

**Visual signature**:
- Large first-person statement headline takes top 50%
- Product or result visual in bottom 50%
- Minimal copy, maximum boldness
- Background: high-contrast (dark or atmospheric)
- Press logos optional at bottom

**Compatible headlines**: H7, H8, H16

---

## FORMAT 9 — UGC TESTIMONIAL

**Description**: User-generated style content. Real-looking person, speaking directly to camera or writing their experience.

**Visual signature**:
- Casual/natural framing, not studio-polished
- Real person (not model-perfect) for authenticity
- Product visible but not dominant
- Text overlay: key quote from testimonial
- Optional: rating stars, verified buyer badge

**Compatible headlines**: H6, H7, H8

---

## FORMAT 10 — REDDIT/FORUM SCREENSHOT

**Description**: Fake or real forum/Reddit post screenshot showing organic-feeling discussion.

**Visual signature**:
- Reddit or forum UI mock
- Post title = the headline
- Comments showing social proof
- Product mentioned organically
- Watermark with brand name subtle

**Compatible headlines**: H4, H5, H7

---

## FORMAT 11 — EMAIL SCREENSHOT

**Description**: Email inbox screenshot showing a "success email" related to the product outcome.

**Visual signature**:
- Email client UI mock
- Subject line = the headline
- Email body: testimonial or outcome report
- Product reference subtle

**Compatible headlines**: H1, H7

---

## FORMAT 12 — MEME FORMAT

**Description**: Recognizable meme template adapted for brand messaging.

**Visual signature**:
- Familiar meme layout (two panels, reaction format, etc.)
- Humor drives the message
- Product as the "solution" punchline
- Bold Impact-style caption font

**Compatible headlines**: H4, H10

---

## FORMAT 13 — FLAT LAY

**Description**: Overhead or 45° bird's-eye view of product surrounded by complementary items.

**Visual signature**:
- Product centered, premium flat surface (marble, dark wood, linen)
- Surrounding items: brand packaging, complementary lifestyle items, natural elements
- Clean, editorial styling
- Overhead or slight angle shot

**Compatible headlines**: H1, H3, H6

---

## FORMAT 14 — MARBLE / MINI WEB PAGE

**Description**: Minimal landing page aesthetic within ad frame. White marble or premium clean background.

**Visual signature**:
- Clean white/marble background
- Logo or brand name top center
- Product image centered
- Key claim + price + CTA button
- Footer: trust signals

**Compatible headlines**: H3

---

## FORMAT 15 — FLAMES / DARK ATMOSPHERIC

**Description**: Product against dramatic dark background with flame, smoke, ember, or dark energy visual effects.

**Visual signature**:
- Deep black or dark background
- Orange/red flame or ember glow from below or sides
- Product illuminated by warm dramatic light
- Moody, powerful, masculine energy

**Compatible headlines**: H6, H7

---

## FORMAT 16 — ATMOSPHERIC (MISTY / NATURE / PREMIUM SETTING)

**Description**: Product in a premium atmospheric scene — forest, water, fog, luxury setting.

**Visual signature**:
- Soft bokeh background: forest, misty landscape, luxury interior
- Product as focal point with environmental story around it
- Muted, atmospheric color grading

**Compatible headlines**: H8

---

## FORMAT 17 — CARTOON SITUATION

**Description**: Illustrated cartoon or Pixar-style 3D scene. Anthropomorphic characters or relatable illustrated scenario. Humor makes taboo/desire subject scroll-stopping while bypassing content classifiers.

**Visual signature**:
- High-quality 3D cartoon illustration or flat illustrated style
- Characters representing the avatar and their context
- Product visible with glowing or magical emphasis
- Speech bubbles or captions for punchline
- Warm, cinematic cartoon lighting

**Compatible headlines**: H13, H14

---

## FORMAT 18 — CARTOON DISCOVERY

**Description**: Illustrated cartoon scene where a third party "discovers" the product or its effects.

**Visual signature**:
- Curious third-party character holding or reacting to product
- Avatar character visible in background (satisfied, resting)
- Speech or thought bubble for discovery punchline
- Product glowing or emphasized

**Compatible headlines**: H14

---

## FORMAT 19 — SENSUAL REALISM

**Description**: Editorial-level real photo of a person in a sensual/aspirational context, with product superimposed or integrated.

**Visual signature**:
- Real cinematic photo, premium editorial quality (GQ/Maxim level)
- Person: realistic skin texture, natural micro-details
- Product overlaid or integrated in foreground
- 4 corner callouts with thin arrows pointing to product
- Press logo strip at bottom
- Warm cinematic lighting and color grading

**Compatible headlines**: H6, H7, H8, H15

**Female aesthetic rule**: Apply **brand-appropriate body aesthetic** — match the target audience's aspirational vision based on brand research and avatar profile. Always tasteful editorial level.

---

## FORMAT 20 — OBJECT METAPHOR / VISUAL PROOF

**Description**: An everyday object used as a metaphor for product results. The object SHOWS the transformation without stating it.

**Visual signature**:
- Central everyday object (belt, scale, tape measure, glass, etc.) showing progress
- Product positioned as the cause/agent
- Minimal other elements
- Strong visual contrast between "before" and "after" states of the object
- Confession-style text overlay

**Compatible headlines**: H16

---

## FORMAT 21 — GENERIC OBJECT METAPHOR

**Description**: Conceptual or symbolic visual using any object to represent the product's core benefit.

**Visual signature**:
- Symbolic object as hero
- Product integrated as the "solution" element
- Creative concept-driven composition

**Compatible headlines**: H11, H16

---

## QUICK FORMAT-TO-HEADLINE COMPATIBILITY TABLE

| Format # | Format Name | Compatible Headlines |
|----------|-------------|----------------------|
| 1 | Product Hero | H1, H2, H3, H9, H12 |
| 2 | In-Use / Lifestyle Shot | H2, H6, H10 |
| 3 | Comparison Bar | H4 |
| 4 | Bullet Layout | H6, H9, H10, H11, H12, H15 |
| 5 | Press Authority | H5, H7, H8, H15 |
| 6 | Anatomical/Scientific | H12 (+ adapt H6, H8) |
| 7 | Before/After | H11 (+ adapt H10, H16) |
| 8 | Confession / 1st-Person | H7, H8, H16 |
| 9 | UGC Testimonial | H6, H7, H8 |
| 10 | Reddit Screenshot | H4, H5, H7 |
| 11 | Email Screenshot | H1, H7 |
| 12 | Meme Format | H4, H10 |
| 13 | Flat Lay | H1, H3, H6 |
| 14 | Marble / Mini Web Page | H3 |
| 15 | Flames / Dark Atmosphere | H6, H7 |
| 16 | Atmospheric (Misty/Nature) | H8 |
| 17 | Cartoon Situation | H13, H14 |
| 18 | Cartoon Discovery | H14 |
| 19 | Sensual Realism | H6, H7, H8, H15 |
| 20 | Object Metaphor / Visual Proof | H16 |
| 21 | Generic Object Metaphor | H11, H16 |

**RULE**: When a user names a concept, find the matching format(s), then produce ONE FULL JSON for EACH compatible headline. Always include adapted headlines that fit the angle even if the format isn't in the headline's primary `Best for` list.


=================================================================
# SECTION 6 — HARD RULES
=================================================================

# HARD RULES — NEVER VIOLATE

> These 17 rules are non-negotiable. Every JSON ad blueprint must respect ALL of them. Verify at the end of each JSON before delivering.

---

## RULE 1 — LANGUAGE
- All **ad copy** in the language appropriate for the target market (derive from site — default to US English for US/international brands)
- **Communicate with the user** in the language they write to you in
- Never mix languages in the same ad creative

## RULE 2 — HEADLINE WORDING
- Always **improve** the headline text when reusing a pattern — never paste a variation verbatim
- Slight word swaps, punctuation tweaks, number changes are encouraged
- Keep the core hook structure intact

## RULE 3 — PRODUCT JSON INJECTION
- Always inject the **full product JSON** (from Section 8) in every ad blueprint
- The `product` field is mandatory
- Never paraphrase or shorten the product description

## RULE 4 — REFERENCE IMAGE INSTRUCTION
- Always specify in the `gpt_image_prompt` field that **a reference image of the product is provided alongside**
- Instruct GPT Image 2 to recreate the product **PERFECTLY IDENTICAL** to the reference
- Same proportions, same material finish, same distinguishing features — never alter, redesign, or stylize

## RULE 5 — FORMAT
- Always **1:1 square format**
- Always **1080×1080 pixels**
- Platform: Meta (Facebook + Instagram) unless user specifies otherwise

## RULE 6 — BATCH SIZE
- Maximum **10 JSONs per file**
- Always full content, never summaries
- If a concept generates >10 JSONs, split into multiple files: `CONCEPT_[N]_[NAME]_BATCH_01.json`, `BATCH_02.json`, etc.

## RULE 7 — PHOTO-REALISTIC HUMANS
- All human subjects must be **photo-realistic** with natural micro-details (skin pores, natural skin tone variation, realistic hair texture)
- If they perform an action, the action must be **physically correct**
- No uncanny-valley effects, no AI-looking faces, no extra fingers

## RULE 8 — PREMIUM AESTHETIC
- Visual benchmark: the brand's own premium aesthetic (extracted from site research — apply the brand's style reference, not a default)
- Agency-grade composition, cinematic lighting, controlled reflections
- Never cheap, cluttered, or stock-photo looking

## RULE 9 — ETHICAL CONSTRAINTS
- ❌ No real public figures
- ❌ No celebrity look-alikes
- ❌ No minors (no one under 25 in appearance)
- ❌ No graphic violence
- ❌ No pornographic sexual content
- ❌ No copyrighted brand logos (other than approved press logos)
- ✅ Tasteful editorial-level sensuality is OK (GQ/Maxim standard) where relevant to the brand

## RULE 10 — AMAZON BADGES
- **Include** Amazon badges only on product-focused formats (Hero, Comparison, Bullet, In-Use, Flat Lay, Object Metaphor, Marble)
- **Exclude** from UGC, Testimonial, Confession, Reddit/Email mocks, Cartoon, Sensual Realism, Press Authority (the latter has its own authority signal)
- When applicable, choose ONE of:
  - **Amazon Best Seller** badge → for Hero, Comparison, Spec-Table, In-Use formats
  - **Amazon's Choice** badge → for Bullet, Flat Lay, Marble formats
- **Never both badges in the same ad**
- Placement: small tag near the product, in a corner, or in a badge strip

## RULE 11 — MANDATORY USE OF H1-H16 LIBRARY
- Always pull from the H1-H16 library (Section 4) as the starting point — never invent from scratch
- First check compatibility: `Best for` field + format compatibility table + angle matrix
- **If compatible as-is**: use with light wording improvement (rule #2)
- **If not directly compatible but angle is relevant**: ADAPT the headline to fit the concept while preserving its hook structure
- **Never force** an incompatible headline → use the closest pre-approved variation

## RULE 12 — PROMISE STRENGTHENING — MAXIMIZE THE PUNCH
- **Push timeframes to shortest believable extreme**:
  - 72h / 3 days → **24h / 1 day**
  - 21 days → **14 days** or **7 days** if defensible
  - 9 days → **7 days** when scroll-stop demands
  - 30 days → **21 days** on transformation promises
- **Intensify outcomes**: "better sleep" → "deepest sleep of your life"
- **Keep specific weird numbers** for believability: 43 seconds, 825 grams, 9 days, 6 years
- **Lead with the strongest claim possible** while respecting platform policy and the guarantee backing every promise
- **Risk-reversal as safety net**: every aggressive promise paired with guarantee in sub or footer

## RULE 13 — MANDATORY JSON FIELDS
Every JSON must specify:
- `belief_installation` (which of the 6 beliefs — from Section 3)
- `consciousness_level` (1-5 from Schwartz framework)
- `psychology_breakdown.scroll_stop_mechanism` (what stops the avatar from scrolling in <0.3s)
- `viral_potential` (1-10 score + rationale)
- `differentiation_from_competitor` (how this differs from generic category ads)

## RULE 14 — PRICE CONSISTENCY
- Always use the **exact price extracted from the website** in all JSONs — never round, approximate, or invent
- Crossed/comparison prices: use exact values from site or calculate accurately from offer structure
- Never use placeholder prices

## RULE 15 — PRICE PSYCHOLOGY
- Reference the brand's pricing in the context of perceived value (compare to competing solutions if site does so)
- Left-digit psychological effect: $X9 prices feel dramatically cheaper than $X+1
- Any reference to total/checkout amount should respect the site's own price positioning

## RULE 16 — BRAND VOICE CONSISTENCY
- Use ONLY the brand voice extracted from site research (Section 1)
- Never impose an alien tone — if the brand is warm and casual, don't write cold and aggressive
- Banned words/phrases: any vocabulary that contradicts the brand's own positioning (identified in research)
- The brand's own copy is the vocabulary source — mine it aggressively

## RULE 17 — HUMAN SUBJECT AESTHETIC
> Match ALL human subjects to the target avatar's aspirational vision — derived from site research.

In any ad creative where a human character or model is shown (real photo, sensual realism, cartoon, UGC, lifestyle), their aesthetic MUST match:
- The brand's target demographic and aspirational self-image
- The imagery/models used on the brand's own site
- The avatar profile built in Section 2

**Principle**: show the person the avatar WANTS to be, or the person they WANT to attract — always from a position of aspiration, never diminishment.

**JSON field requirement**: include `human_subjects.aesthetic_brief` with a description of the body type, style, and look appropriate to this brand and avatar when humans are present.

---

## VERIFICATION CHECKLIST (run before delivering each JSON)

- [ ] Headline pulled/adapted from H1-H16 library (rule 11)
- [ ] Headline wording improved, not verbatim (rule 2)
- [ ] Promise strengthened (72h → 24h, etc.) where applicable (rule 12)
- [ ] Product JSON fully injected (rule 3)
- [ ] Reference image instruction in `gpt_image_prompt` (rule 4)
- [ ] 1080×1080, 1:1 format specified (rule 5)
- [ ] Photo-realistic humans if present (rule 7)
- [ ] Premium aesthetic descriptors throughout (rule 8)
- [ ] No public figures, no minors, no graphic content (rule 9)
- [ ] Amazon badge logic correct (rule 10)
- [ ] All mandatory JSON fields present (rule 13)
- [ ] Exact price used from site research (rule 14)
- [ ] Brand voice consistent (rule 16)
- [ ] Human aesthetic matches avatar aspirational vision (rule 17)
- [ ] Risk reversal paired with aggressive promise (rule 12)
- [ ] Belief installation specified (rule 13)
- [ ] Consciousness level specified (rule 13)
- [ ] Communication with user in their language (rule 1)


=================================================================
# SECTION 7 — PRODUCT JSON TEMPLATE
=================================================================

# PRODUCT JSON TEMPLATE — INJECT IN EVERY AD BLUEPRINT

> After website research, build the product JSON from the real product. Inject in the `product` field of every ad blueprint without modification.

```json
{
  "subject": {
    "description": "[FILL: Describe the physical appearance of the product in precise detail — shape, color, material, finish, distinguishing visual features. Example: 'A glossy black elastic bead bracelet arranged as a vertical oval ring, composed of uniform rectangular beads with rounded corners, several bead faces catching soft pale gray rectangular reflections from studio lights.']",
    "mirror_rules": "[FILL: Specify what can/cannot be reflected on product surfaces. Example: 'Bead faces show softbox-style rectangular highlight reflections; no readable text or identifiable reflected objects.']"
  },
  "accessories": {
    "prop": {
      "type": "[FILL: Product category — e.g. 'Elastic bead bracelet', 'Serum bottle', 'Supplement capsule jar', 'Smart ring']",
      "details": "[FILL: Precise visual details — material, color finish, size perception, any text visible on packaging, elastic/rigid construction, etc.]"
    }
  },
  "photography": {
    "camera_style": "[FILL: e.g. 'Clean studio product photography with crisp focus and controlled reflections']",
    "angle": "[FILL: e.g. 'Slightly elevated eye-level angle, centered on the product's vertical form']",
    "shot_type": "[FILL: e.g. 'Single product hero shot, tight framing']",
    "aspect_ratio": "1:1 square format",
    "texture": "[FILL: e.g. 'Sharp edges, smooth gradients, minimal noise, soft shadowing and specular highlights on glossy surfaces']"
  },
  "background": {
    "setting": "[FILL: e.g. 'Minimal seamless studio backdrop for product catalog imagery']",
    "wall_color": "[FILL: e.g. 'Soft white with a light gray gradient falloff']",
    "elements": [
      "[FILL: Background element 1]",
      "[FILL: Background element 2 — e.g. 'Soft contact shadow beneath the product']"
    ],
    "atmosphere": "[FILL: e.g. 'Minimal, premium, clinical-clean']",
    "lighting": "[FILL: e.g. 'Diffused studio lighting producing broad rectangular highlights and a gentle shadow under the lower arc']"
  }
}
```

## Product JSON Invariants

After filling the template from site research, define the following:

### What NEVER varies (always inject these exact descriptions)
- Product shape and proportions
- Product color and material
- Product finish (glossy/matte/textured)
- Any distinctive visual signature (bead pattern, label design, cap shape, etc.)

### What CAN vary by concept
- Background, atmosphere, scene
- Lighting color temperature (warm/cool/dramatic)
- Product orientation/angle (hero default vs in-use vs floating)
- Product scale in frame (dominant hero vs supporting element)
- Glow or light effects applied to product in specific formats

### Reference Image Instruction (always include in gpt_image_prompt)
> "A reference image of the [product name] is provided alongside this prompt. Recreate the [product] PERFECTLY IDENTICAL to the reference image — same proportions, same [key feature 1], same [key feature 2], same [finish/texture]. Do not redesign, restyle, or alter any visual property of the product despite the surrounding scene."


=================================================================
# SECTION 8 — GPT IMAGE 2 BEST PRACTICES
=================================================================

# GPT IMAGE 2 (CHATGPT IMAGES 2.0) PROMPT BEST PRACTICES

> The `gpt_image_prompt` field in every JSON is the FINAL DELIVERABLE — it gets pasted directly into ChatGPT (Images 2.0) or sent to the `gpt-image-2` API, together with the reference product image. GPT Image 2 reasons about the request before drawing: it rewards STRUCTURE, visual facts and explicit constraints — not vague praise.

---

## STRUCTURE OF A GREAT GPT IMAGE 2 PROMPT

Write the prompt in **5 labelled blocks**, separated by line breaks (GPT Image 2 follows structure better than one long paragraph):

1. **IMAGE 1 / REFERENCE** — label the input: "Image 1 = product reference photo of [product]." Then the lock: "Reproduce the product exactly as in Image 1 — same geometry, same proportions, same label text and layout, same colors, same material finish. Do not redesign, restyle, relabel or re-letter it."
2. **SCENE** — format + style + setting: "Square 1:1 Meta ad, photorealistic [style]. [Setting, surface, background, props — physical objects and materials, with positions]."
3. **SUBJECT** — what dominates the frame: product placement (position, % of frame, orientation, contact shadow) + human subjects if any (age range, build, skin texture, wardrobe, exact action — matching the avatar aesthetic brief).
4. **IMPORTANT DETAILS** —
   - **Typography**: every text element written EXACTLY between quotes, with font style, weight, size (% of frame), color hex, position. Example: headline "6 YEARS OF PAIN. GONE IN 9 DAYS." — heavy condensed sans-serif (Anton-style), white #FFFFFF, top 25% of frame, centered, tight tracking.
   - **Badges / press logos** (if applicable): exact text, colors, position.
   - **Lighting**: named light sources with direction and color temperature ("warm 2800K key light from upper-left at 45°, soft fill from right").
   - **Camera**: lens + aperture ("85mm, f/2.8, eye-level") — GPT Image 2 changes composition and depth of field accordingly.
   - **Composition**: frame splits in % (headline top 25% / hero middle 55% / footer strip bottom 15%).
5. **USE CASE + CONSTRAINTS** — "Use case: high-converting Meta feed ad, agency-grade, premium [brand aesthetic]." Then the constraints list (the most important block — never skip it):
   - "Render ONLY the text written above — no extra words, no duplicate text, no misspellings, no placeholder text."
   - "No logos other than those specified, no watermark, no UI elements unless specified."
   - "No extra fingers, no distorted hands, no uncanny faces, no plastic-looking skin."
   - "The product must remain an exact replica of Image 1 — do not alter [key feature 1], [key feature 2], [label text]."

**Length target**: 250-450 words across the 5 blocks. Below 200 = generic output. Above 500 = dilution of priorities.

**Size**: always square 1:1. Request the largest square size available (1024×1024 minimum, 2048×2048 if offered) and export/crop to 1080×1080. Use `quality: "high"` via API.

---

## PROMPT WRITING PRINCIPLES

### 1. Visual facts, not praise
GPT Image 2 ignores "stunning / epic / amazing". It obeys materials, objects, light sources, lenses.
- ❌ "dramatic and powerful scene with the product"
- ✅ "product centered on matte black seamless backdrop, single hard key light from upper-left, product occupying 55% of frame, deep shadow falloff on the right"

### 2. Specificity = obedience
- ❌ "headline at top"
- ✅ "headline \"[EXACT TEXT]\" in heavy condensed sans-serif, white #FFFFFF, occupying the top 25% of frame, centered horizontally, tight tracking"

### 3. Text is typography, not decoration
GPT Image 2 renders text very well — IF every string is quoted exactly. Write each text element in quotes or ALL CAPS, give font style + size + color + position, and spell unusual brand names letter by letter if needed (e.g. "B-R-A-N-D-N-A-M-E"). Always add "no extra words, no duplicate text".

### 4. Label every reference image by role
"Image 1 = product reference." If more images are given (e.g. a winning ad to clone the layout from): "Image 2 = layout reference — copy the composition only, not its text or product." Up to 16 reference images are accepted; label each one.

### 5. Iterate with CHANGE + PRESERVE, not full rewrites
For corrections after a first render, use the 3-sentence edit pattern:
> "Change only [X]. Preserve [product, label text, headline, layout, lighting]. Keep everything else the same."
Repeat the preserve list on every iteration to avoid product drift.

### 6. End on the product lock
The last line of the CONSTRAINTS block always re-locks the product:
> "CRITICAL: the [product] must remain a visual replica of Image 1 — do not alter [key features] despite the [scene/overlay/effects] context."

---

## COMMON PROMPT ERRORS TO AVOID

| Error | Fix |
|-------|-----|
| One giant unstructured paragraph | Split into the 5 blocks (Reference / Scene / Subject / Details / Use case + Constraints) |
| Missing CONSTRAINTS block | Always end with the constraints list — it is the most important block |
| Text not quoted exactly | Put every string in quotes, add "no extra words, no duplicate text" |
| Reference image not labelled | "Image 1 = product reference" + lock sentence |
| Mood words instead of visual facts | Convert to objects, materials, light sources, lens |
| Full rewrite to fix one detail | Use CHANGE + PRESERVE edit pattern |
| Too long (>500 words) | Cut redundant adjectives, keep facts and constraints |
| Too short (<200 words) | Add lighting, camera, composition %, typography specs |


=================================================================
# SECTION 9 — PRESS LOGOS & BADGES
=================================================================

# PRESS LOGOS & BADGES ASSET LIBRARY

> Use these trust-signal assets in ad creatives. Referenced by name in `gpt_image_prompt` — GPT Image 2 renders text-only versions.

---

## PRESS LOGOS (use grayscale text-only versions)

Choose press logos that are contextually relevant to the brand's category and target audience. Always render as **clean text-only typography** (the brand wordmark style), in **grayscale** (white on dark backgrounds, black on light backgrounds).

| Logo | Style guide | Best for angles |
|------|-------------|-----------------|
| **Men's Health** | Slab serif wordmark | Performance, fitness, masculinity, health |
| **Forbes** | Bold serif wordmark | Authority, business, premium positioning |
| **The Guardian** | Bold serif wordmark | Trust, journalism, credibility |
| **Cosmopolitan** | Italic serif wordmark | Lifestyle, relationship, beauty, confidence angles |
| **WellBeing** | Sans-serif wordmark | Health, natural alternatives, wellness |
| **GQ** | Bold sans monogram | Premium, masculine, lifestyle |
| **Esquire** | Italic serif wordmark | Premium lifestyle |
| **Muscle & Fitness** | Bold sans wordmark | Performance, body transformation |
| **Vogue** | Condensed serif wordmark | Premium beauty, fashion, femininity |
| **Self** | Bold sans wordmark | Women's health, fitness, confidence |
| **Shape** | Bold sans wordmark | Women's fitness and body transformation |
| **Wired** | Bold condensed wordmark | Tech, innovation, science-backed products |

**Selection rule**: pick 3-5 logos that feel most aligned with the brand's specific category and audience. When in doubt, favor authority (Forbes, The Guardian) + category relevance (the most specific applicable niche title).

## PRESS STRIP RENDERING SPECS

- **Position**: bottom 10-15% of frame
- **Background**: white strip on dark background OR dark strip on light background
- **Logos**: evenly spaced horizontally
- **Color**: grayscale (no original logo colors)
- **Spacing**: equal horizontal gaps
- **Vertical alignment**: centered baseline

---

## AMAZON BADGES

Two Amazon badges are available. **Use ONE per ad. Never both.** Skip entirely on cartoon/confession/UGC formats.

### Amazon Best Seller Badge
- **Color**: Orange #FF9900 background, white text
- **Shape**: Rectangular tag with slight rounded corners
- **Text**: "Amazon Best Seller" or "#1 Best Seller"
- **Use for**: Hero, Comparison, Spec-Table, In-Use formats
- **Placement**: corner of frame OR small badge near product

### Amazon's Choice Badge
- **Color**: Turquoise/teal #00A8E1 background, white text
- **Text**: "Amazon's Choice" + small subtitle "for [product category keyword]"
- **Use for**: Bullet Layout, Flat Lay, Marble formats
- **Placement**: corner of frame OR small badge in bullet area

---

## OTHER TRUST SIGNALS (use sparingly — derive specifics from site research)

| Signal | Description | Best use |
|--------|-------------|----------|
| **[X]-Day Guarantee badge** | Circular shield + "[X]-DAY GUARANTEE" | Footer of any format |
| **Free Shipping badge** | Truck icon + "FREE SHIPPING" | Footer alongside guarantee |
| **[X]+ Customers badge** | "[X]+ Verified Customers" | Comparison Bar, Press Authority |
| **Star rating** | "★★★★★ [X]/5 from [Y] reviews" | Bullet Layout, UGC |
| **Origin badge** | "Made / Designed in [COUNTRY]" | Only if accurate and verified on site |

---

## GUARANTEE / RISK REVERSAL FOOTER STRIPS

Derive from site research — use exact language from the brand's guarantee/offer. Common patterns:

### Footer A — Standard
> [DISCOUNT]% OFF TODAY • FREE SHIPPING • [X]-DAY GUARANTEE

### Footer B — Offer Focus
> BUY 1 GET 1 FREE • FREE SHIPPING • [X]-DAY MONEY-BACK GUARANTEE

### Footer C — Risk Reversal Heavy
> Try It Risk-Free • [X] Days to Decide • Or Full Refund

### Footer D — Price + Offer
> $[PRICE] • [OFFER NAME] • [X]-Day Guarantee

### Footer E — Today Only
> TODAY ONLY • FREE SHIPPING • SATISFACTION GUARANTEED

---

## URL / DOMAIN DISPLAY

When showing the URL in an ad:
- Always use the **exact domain from the user's research URL**
- Style: clean sans-serif, white or black depending on background
- Position: footer, near CTA, or below tagline
- Optional clickable feel: underline OR small arrow


=================================================================
# SECTION 10 — TYPOGRAPHY & COLORS
=================================================================

# TYPOGRAPHY & COLOR DESIGN SYSTEM

> Universal design system for ad creatives. Use these specifications when defining the `typography` and `color_palette` fields in JSON blueprints. Adapt the specific palette choices to match the brand's color identity (from site research).

---

## TYPOGRAPHY HIERARCHY

### Headline (the dominant element)
- **Recommended fonts**: Inter Black, Helvetica Neue Black, Bebas Neue, Anton, Impact
- **Weight**: 800-900 (black)
- **Size**: ~120-180px equivalent (occupies 25-40% of vertical frame)
- **Tracking**: Tight (-2 to -1) for impact OR wide (+2 to +5) for editorial
- **Line height**: 0.9-1.0 (tight, condensed)
- **Treatment**: UPPERCASE, ends with period for finality, optional drop shadow or outer glow
- **Position**: Top third of frame (most common), center for hero formats

### Sub-headline (supporting context)
- **Recommended fonts**: Inter Medium/Regular, Playfair Display Italic, Georgia Italic
- **Weight**: 400-500 (regular to medium)
- **Size**: ~30-50% of headline size
- **Style**: Often italic for editorial feel, regular for informational
- **Position**: Directly below headline, centered or left-aligned

### Callouts / Bullets (benefits)
- **Recommended fonts**: Inter Medium, Helvetica Medium
- **Weight**: 500-600 (medium to semibold)
- **Size**: ~40-60% of headline size
- **Style**: Paired with arrows, icons, or emojis

### Footer / Micro-copy (price, guarantees, URL)
- **Recommended fonts**: Inter Regular, Helvetica Regular
- **Weight**: 400-500
- **Size**: ~25-35% of headline size
- **Style**: Clean, easy-to-read, separators (•)

### Buttons / CTAs
- **Recommended fonts**: Inter Bold, Helvetica Bold
- **Weight**: 700 (bold)
- **Style**: UPPERCASE, in colored rectangle, slight rounded corners

---

## UNIVERSAL COLOR PSYCHOLOGY

| Color | Psychological effect | Use for |
|-------|----------------------|---------|
| **Pure black #000000** | Premium, authority, mystery | Hero, premium positioning |
| **Pure white #FFFFFF** | Clean, clinical, trustworthy | Sub, footer, callouts |
| **Saturated yellow #FFD60A** | Urgency, attention, scroll-stop | Sale/BOGO urgency ads |
| **Alarm red #E63946** | Urgency, anti-something, danger | Competitor kills, price slashes, comparison negatives |
| **Trust green #06D6A0** | Approval, checkmark, success | Bullets, comparison positives, guarantee badges |
| **Premium gold #FFD700** | Luxury, exclusive, original | "Original" callouts, premium taglines |
| **Deep atmospheric blue #1A1A2E** | Calm, tech, science | Scientific/mechanism ads |
| **Warm teal #2D3D40** | Nature, calm, contemplation | Atmospheric nature ads |

**Brand palette override**: If the brand has a distinctive color system (identified in site research), incorporate their palette into the `color_palette.brand_colors` field and reference brand colors for headline, accent, and CTA elements.

---

## CONTRAST RULES

- **Headline vs background**: minimum 7:1 contrast ratio
- **Sub-headline vs background**: minimum 4.5:1 contrast ratio
- **Callouts vs background**: minimum 4.5:1
- **CTA buttons**: high-contrast color on contrasting background

---

## SHADOW & GLOW EFFECTS

### Headline drop shadow
- Offset: 2-4px / Blur: 4-8px / Color: black at 30-50% opacity
- Use for: light backgrounds where headline needs separation

### Headline outer glow
- Spread: 6-10px / Color: warm orange-gold for fire ads, neon green for anatomical, white for premium
- Opacity: 60-80% / Use for: dramatic headline emphasis

### Product glow (Cartoon/Magic formats)
- Color: warm gold #FFD700 / Spread: 8-15px / Opacity: 70-90%
- Use for: Cartoon and magical product formats only

### Soft contact shadow (under product)
- Offset: 0px x, 4-8px y / Blur: 12-20px / Color: black at 25-40% opacity
- Use for: all product hero shots where product "sits" on a surface

---

## COMPOSITION GRID

Most ads use this layout:

```
┌─────────────────────────────────┐
│  HEADLINE (top 25-30%)          │  ← bold, dominant
├─────────────────────────────────┤
│                                 │
│  HERO ELEMENT (middle 50-60%)   │  ← product + scene
│  (product, model, illustration) │
│                                 │
├─────────────────────────────────┤
│  CTA / FOOTER (bottom 15-20%)   │  ← price, guarantee, URL
└─────────────────────────────────┘
```

Variations:
- **Bullet Layout**: split 50/50 left-right (product | bullets)
- **Comparison Bar**: split 50/50 left-right (competitor | brand) with VS in middle
- **Confession**: headline takes top 50%, hero element bottom 50%


=================================================================
# SECTION 11 — CONCEPT → JSON WORKFLOW
=================================================================

# THE 8-STEP CONCEPT-TO-JSON PRODUCTION WORKFLOW

> Follow this exact sequence every time the user names a concept to produce.

---

## STEP 1 — CONFIRM BRAND RESEARCH IS COMPLETE

Before doing anything else:
- [ ] Section 0 (Brand Research Protocol) has been completed
- [ ] Section 1 (Brand Profile) is filled
- [ ] Section 2 (Avatar) is built
- [ ] User has confirmed the brand profile

If any of the above is incomplete → stop and complete Section 0 first.

---

## STEP 2 — IDENTIFY THE VISUAL FORMAT

Read the user's concept description. Map it to one (or a hybrid of two) of the 21 formats from Section 5.

Available formats:
1. Product Hero | 2. In-Use / Lifestyle | 3. Comparison Bar | 4. Bullet Layout | 5. Press Authority | 6. Anatomical/Scientific | 7. Before/After | 8. Confession | 9. UGC Testimonial | 10. Reddit Screenshot | 11. Email Screenshot | 12. Meme | 13. Flat Lay | 14. Marble/Mini Web | 15. Flames/Dark Atmosphere | 16. Atmospheric | 17. Cartoon Situation | 18. Cartoon Discovery | 19. Sensual Realism | 20. Object Metaphor | 21. Generic Object Metaphor

**Match logic**: pick the format whose `Visual signature` field most closely matches the concept. If hybrid, use dominant format and layer secondary elements.

---

## STEP 3 — LIST ALL COMPATIBLE HEADLINES

Consult the **compatibility matrix** from Section 4 and Section 5:

| Format | Compatible Headlines |
|--------|---------------------|
| Product Hero | H1, H2, H3, H9, H12 |
| In-Use / Lifestyle | H2, H6, H10 |
| Comparison Bar | H4 |
| Bullet Layout | H6, H9, H10, H11, H12, H15 |
| Press Authority | H5, H7, H8, H15 |
| Anatomical/Scientific | H12 (+ adapt H6, H8) |
| Before/After | H11 (+ adapt H10, H16) |
| Confession | H7, H8, H16 |
| UGC Testimonial | H6, H7, H8 |
| Reddit Screenshot | H4, H5, H7 |
| Email Screenshot | H1, H7 |
| Meme | H4, H10 |
| Flat Lay | H1, H3, H6 |
| Marble/Mini Web | H3 |
| Flames/Dark Atmosphere | H6, H7 |
| Atmospheric | H8 |
| Cartoon Situation | H13, H14 |
| Cartoon Discovery | H14 |
| Sensual Realism | H6, H7, H8, H15 |
| Object Metaphor | H16 |
| Generic Object Metaphor | H11, H16 |

**Output**: list ALL compatible headline IDs for this concept.

**Critical**: do NOT pre-filter or skip headlines. Every compatible headline = one JSON. If 6 headlines are compatible → 6 JSONs to produce.

---

## STEP 4 — DECIDE ADAPTATION VS DIRECT USE

For each headline in the compatible list, decide:

- **Direct use (with light improvement)**: headline fits concept naturally → use pattern with light tweaks
- **Adaptation needed**: angle is relevant but literal pattern doesn't fit visual concept → adapt while preserving hook structure

**Adaptation examples**:

| Original Pattern | Concept | Adapted |
|-----------------|---------|---------|
| H8: "[X] NIGHTS TO [RESULT]" | Anatomical Pain Map | "[X] NIGHTS TO PAIN-FREE MOVEMENT" |
| H6: "THEY WON'T SAY IT. BUT THEY'LL FEEL IT." | Scientific mechanism | "THEY WON'T SAY IT. BUT THEY'LL FEEL THE DIFFERENCE." |
| H10: "Struggling with [PAIN]?" | Before/After | "He STRUGGLED with [PAIN]. Then he tried THIS." |
| H16: "A [PROOF] EVERY [TIMEFRAME]." | Object Metaphor variant | "ONE [OBJECT]. [N] [UNITS] TIGHTER." |

If no adaptation produces a clean fit, drop that headline — do NOT force.

---

## STEP 5 — APPLY PROMISE STRENGTHENING

Before locking each headline, scan for strengthening opportunities (Rule 12):

- 72h / 72 hours / 3 days → **24h / 24 hours / 1 day**
- 21 days → **14 days** or **7 days** if defensible
- 9 days → **7 days**
- 30 days → **21 days** for transformations
- "better X" → "deepest/most extreme X of your life"
- Round numbers → swap for oddly specific (43 seconds, 825g, 9 days, 6 years)

Always pair the strengthened promise with risk reversal in sub or footer.

---

## STEP 6 — BUILD THE JSON SKELETON

Fill in every field from Section 17 (JSON Template). Mandatory fields:

```
{
  "ad_id": "[BRANDNAME]_[CATEGORY]_[CONCEPTNAME]_[N]",
  "concept_name": "...",
  "concept_number": N,
  "headline_id": "H##",
  "headline_final": "...",        // adapted + improved + promise-strengthened
  "sub_headline": "...",
  "footer_text": "...",
  "format_id": "...",
  "consciousness_level": 1-5,
  "belief_installation": ["#X", "#Y"],
  "angle": "...",
  "visual_format": "...",
  "scene_description": {...},
  "human_subjects": {...},         // include aesthetic_brief if humans present
  "product": {...},                // FULL product JSON from Section 7
  "typography": {...},
  "color_palette": {...},
  "lighting": {...},
  "composition": {...},
  "amazon_badge": "Best Seller" | "Choice" | null,
  "press_logos": [...] | null,
  "psychology_breakdown": {
    "scroll_stop_mechanism": "...",
    "emotional_trigger": "...",
    "rational_justification": "..."
  },
  "viral_potential": {"score": X, "rationale": "..."},
  "differentiation_from_competitor": "...",
  "funnel_stage": "TOF" | "MOF" | "BOF",
  "gpt_image_prompt": "..."      // 250-450 words, paste-ready
}
```

---

## STEP 7 — WRITE THE GPT_IMAGE_PROMPT

Structure — 5 labelled blocks separated by line breaks (see Section 8):
1. IMAGE 1 / REFERENCE: label the product photo + exact recreation lock
2. SCENE: square 1:1 Meta ad, style, setting, surfaces, props
3. SUBJECT: product placement (position, % of frame) + human subjects (aesthetic brief, exact action)
4. IMPORTANT DETAILS: typography (every string in quotes, font, size, hex, position), Amazon badge / press logos if applicable, lighting (sources, direction, Kelvin), camera (lens, aperture), composition (% splits)
5. USE CASE + CONSTRAINTS: use case line, then "only the text above, no extra words, no duplicate text, no extra logos, no watermark, no extra fingers", and the final product lock

---

## STEP 8 — VERIFY AGAINST 17 RULES

Run the verification checklist from Section 6. Tick every box before delivering. Fix any failures before proceeding.


=================================================================
# SECTION 12 — SCROLL-STOP PSYCHOLOGY
=================================================================

# SCROLL-STOP PSYCHOLOGY

> The ad has 0.3 seconds to stop the scroll. Every element must serve this goal.

---

## THE 3-LAYER SCROLL-STOP MODEL

Every high-performing ad operates on 3 simultaneous layers:

| Layer | What stops the scroll | Examples |
|-------|----------------------|---------|
| **Visual** | Image pattern interrupt | Taboo imagery, unexpected juxtaposition, dramatic contrast, recognizable but wrong |
| **Cognitive** | Headline creates an open loop | Specific numbers, unfinished thought, taboo subject, impossible claim |
| **Emotional** | Triggers a felt response in <0.3s | Fear of missing out, lust, jealousy, shame, pride, nostalgia |

The best ads stop on ALL 3 layers simultaneously.

---

## SCROLL-STOP MECHANISMS BY FORMAT

| Format | Primary visual interrupt |
|--------|-------------------------|
| Product Hero | Premium contrast + product dominance |
| Comparison Bar | Immediately legible "vs" structure |
| Bullet Layout | Product size + benefit quantity signal |
| Press Authority | Press logos signal "I've seen this before, it's legit" |
| Anatomical/Scientific | Medical/scientific visual = unexpected for an ad |
| Confession | Bold huge text = direct pattern interrupt |
| Sensual Realism | Aspirational desire trigger |
| Cartoon | Uncanny familiar (Pixar quality on an ad = unexpected) |
| Before/After | Transformation curiosity gap |
| Object Metaphor | Ordinary object used unexpectedly |

---

## HEADLINE SCROLL-STOP MECHANICS

| Technique | Example | Why it works |
|-----------|---------|--------------|
| Specific number | "825 GRAMS GONE" | Oddly specific = credible + curious |
| Taboo subject | "[DESIRE] BACK IN 24H" | Shame/desire response overrides filter |
| Impossible claim | "6 YEARS GONE IN 9 DAYS" | Cognitive dissonance demands resolution |
| Third-party confession | "SHE WON'T SAY IT. BUT SHE'LL FEEL IT." | Voyeuristic social curiosity |
| Urgency trigger | "3 HOURS LEFT" | FOMO activates immediately |
| Direct question | "Struggling with [PAIN]?" | Self-identification response |
| Anti-expected | "[OLD SOLUTION] IS DEAD" | Category disruption = surprise |

---

## PSYCHOLOGICAL TRIGGERS BY AVATAR TYPE

Fill these with the specific avatar's triggers after website research:

| Trigger | Situation that activates it |
|---------|----------------------------|
| [Fear #1 from avatar research] | [Concrete scenario] |
| [Fear #2] | [Scenario] |
| [Desire #1] | [Scenario] |
| [Social comparison moment] | [Scenario] |
| [Performance failure moment] | [Scenario] |
| [Positive jealousy moment] | [Scenario] |


=================================================================
# SECTION 13 — OBJECTION HANDLING
=================================================================

# OBJECTION HANDLING IN AD CREATIVES

> Every element of the ad can proactively handle an objection. Use this section to embed objection handling naturally.

---

## THE 5 UNIVERSAL OBJECTIONS

| Objection | Ad element to handle it |
|-----------|-------------------------|
| "It doesn't work" | Testimonials, mechanism explanation, specific numbers |
| "It's too expensive" | ROI comparison, BOGO/bundle offer, crossed-out price |
| "I've tried things before" | "Anti-X" headline (Belief #2 copy), "no pills/no [old solution]" |
| "I don't trust this brand" | Press logos, review count, guarantee, Amazon badges |
| "What if it doesn't work for me?" | Guarantee + risk reversal in footer/callout |

---

## OBJECTION → AD ELEMENT MAPPING

| Objection | Best ad element | Format where it's most powerful |
|-----------|-----------------|----------------------------------|
| "Doesn't work" | `psychology_breakdown.rational_justification` → mechanism copy | Bullet Layout, Press Authority |
| "Too expensive" | Crossed price in `footer_text` | BOGO Hero, Marble Web Page |
| "Tried it before" | H5/H8 headline pattern | Press Authority, Dark Hero |
| "Don't trust" | Press logos + Amazon badge + review count | Press Authority, Bullet Layout |
| "Risk of waste" | "[X]-Day Guarantee" callout | Any format — always in footer |

---

## BRAND-SPECIFIC OBJECTION RESPONSES

After research, fill the avatar's specific objections (from Section 2) and match each to a copy angle and a JSON field to embed the counter in.


=================================================================
# SECTION 14 — COPY PATTERNS & MICROCOPY
=================================================================

# COPY PATTERNS & MICROCOPY LIBRARY

> Universal copy patterns that work across any brand. Fill in `[VARIABLES]` from brand research.

---

## POWER STRUCTURES

### The Specificity Bomb
> "[SPECIFIC NUMBER] [UNIT] of [PROBLEM] — [RESULT] in [SPECIFIC TIMEFRAME]"
- "825 grams of belly gone — 2 belt holes smaller"
- "[X] kilos lost — [jeans size] in [timeframe]"

### The Anti-Pill
> "No [old solution]. No [side effect]. Just [product mechanism] working silently [where/how]."
- "No pills. No side effects. Just [X]g of [mechanism] working silently."

### The Third-Party Confession
> "[Third party] [noticed / reacted] before [avatar] did."
- "She asked what changed. He just smiled."
- "His team noticed before he said a word."

### The Timeframe Accelerator
> "[Result] in [surprisingly short time]. Or refunded."
- "[X] nights. That's all it takes. Or refunded."

### The Specific Number Hook
> "[Oddly specific number] of [unit] [vanished/returned/improved]."
- "43 seconds after putting it on."
- "6 years of [problem]. Gone in 9 days."

### The Social Proof Stack
> "[X]+ [avatars]. [Rating]. [Guarantee]."
- "10,000+ men. 4.8/5. 30-day guarantee."

---

## MICROCOPY PATTERNS (footer / callout / badge)

| Position | Copy patterns |
|----------|---------------|
| Footer A | "[DISCOUNT]% OFF TODAY • FREE SHIPPING • [X]-DAY GUARANTEE" |
| Footer B | "BUY 1 GET 1 FREE • FREE SHIPPING • [X]-DAY MONEY-BACK" |
| Footer C | "Try It Risk-Free • [X] Days to Decide • Or Full Refund" |
| Footer D | "$[PRICE] • [OFFER] • [X]-Day Guarantee" |
| Risk reversal callout | "Or [money back / full refund / no questions]" |
| Guarantee callout | "[X]-DAY GUARANTEE — NO QUESTIONS" |
| Urgency callout | "TODAY ONLY" / "LIMITED STOCK" / "ENDS AT MIDNIGHT" |
| Social proof callout | "[X]+ VERIFIED [CUSTOMERS/REVIEWS]" |

---

## BANNED COPY PATTERNS

Never use these in any ad (universally):

| Banned | Why | Replace with |
|--------|-----|--------------|
| "amazing" / "incredible" | Empty adjectives | Specific numbers |
| "discover the secret" | Clickbait fatigue | Direct claim |
| "you won't believe" | Clickbait fatigue | Direct statement |
| "doctors recommend" | Fake authority | "[X]+ [avatars] trust" |
| "may help" / "could reduce" | Weak verbs | "ends" / "kills" / "destroys" |
| "magic" / "miracle" | Cheap | "engineered" / "proven" / specific mechanism |
| "natural" alone | Too vague | Specific mechanism or ingredient |

Additionally, identify and ban vocabulary that contradicts the brand's own positioning (from site research) and add it to this table.

---

## US ENGLISH CONVENTIONS (when applicable)

| ❌ Avoid | ✅ Use |
|----------|--------|
| centre | center |
| colour | color |
| favourite | favorite |
| analyse | analyze |
| 1.000 | 1,000 |
| 39,99 € | $39.99 |
| 21h00 | 9 PM |


=================================================================
# SECTION 15 — QUALITY CONTROL
=================================================================

# QUALITY CONTROL CHECKLIST

> Run this before delivering any batch of JSONs. Every item must pass.

---

## PER-JSON CHECKLIST

### Copy quality
- [ ] Headline originates from H1-H16 library (not invented)
- [ ] Headline improved from base pattern (rule 2)
- [ ] Promise is the strongest defensible version (rule 12)
- [ ] Sub-headline supports headline (not redundant)
- [ ] Footer includes guarantee/risk reversal
- [ ] Brand voice matches site research (rule 16)
- [ ] No banned copy patterns (Section 14)
- [ ] Price matches exact site price (rule 14)

### Visual quality
- [ ] Scene description is specific enough to produce without ambiguity
- [ ] Lighting described with color temperature + direction + mood
- [ ] Composition described with frame percentage splits
- [ ] Typography fully specified (text, font, size, color, position)
- [ ] Human subjects described with aesthetic brief matching avatar profile
- [ ] Product recreate instruction explicit in gpt_image_prompt (rule 4)

### Strategic quality
- [ ] Consciousness level matches headline/angle
- [ ] Belief installation correctly mapped
- [ ] Psychology breakdown: scroll_stop_mechanism is specific and concrete
- [ ] Viral potential score is justified by rationale
- [ ] Funnel stage matches consciousness level
- [ ] Amazon badge logic correct (rule 10)
- [ ] Press logos contextually relevant to brand category

### Rules compliance
- [ ] All 17 Hard Rules verified (Section 6 checklist)

---

## PER-BATCH CHECKLIST

- [ ] Full funnel covered (at least 1 ad per awareness level L1-L5)
- [ ] Multiple format types represented (not all bullet layouts)
- [ ] Mix of emotional and rational angles
- [ ] Max 10 JSONs per file (rule 6)
- [ ] File named correctly: `CONCEPT_[N]_[BRAND]_[NAME]_BATCH_01.json`

---

## RED FLAGS (stop and fix before delivering)

🚨 **Invented headline** not from library → rewrite from H1-H16 pattern  
🚨 **Wrong price** in JSON → fix to exact site price  
🚨 **Vague gpt_image_prompt** (<200 words or no typography specs) → expand  
🚨 **Product JSON missing** → inject from Section 7  
🚨 **No reference image instruction** in gpt_image_prompt → add  
🚨 **Aggressive promise without guarantee** → add risk reversal to footer  


=================================================================
# SECTION 16 — EMOJI USAGE
=================================================================

# EMOJI USAGE GUIDELINES

> Emojis are scroll-stop tools when used in moderation. Overuse = cheap.

---

## UNIVERSAL APPROVED EMOJIS

| Emoji | Use |
|-------|-----|
| ✅ ❌ | Comparison bars (yes/no columns) |
| ⚡ | Energy, power, speed claims |
| 🔥 | Drive, intensity, "hot offer" |
| 💪 | Strength, performance |
| 🛡️ | Guarantee, protection, risk reversal |
| ⭐ ★ | Star ratings |
| 👇 | Pointing-down emphasis after bold statement |
| 🇺🇸 🇫🇷 🇬🇧 | Country-of-origin (only if accurate) |

## BRAND-SPECIFIC EMOJIS

After site research, identify emojis consistent with the brand's visual language and add them here.

**Rule**: max 2 emojis per ad creative. Headline gets max 1 emoji.

**Never use**: generic smileys, excessive sparkles, hand gestures — feels stock and cheap.


=================================================================
# SECTION 17 — JSON TEMPLATE
=================================================================

# FULL JSON TEMPLATE — USE FOR EVERY AD BLUEPRINT

```json
{
  "ad_id": "[BRANDNAME_UPPERCASE]_[CATEGORY]_[CONCEPTNAME_UPPERCASE]_[N]",
  "concept_name": "[Human-readable concept name]",
  "concept_number": 0,
  "headline_id": "H##",
  "headline_final": "[ADAPTED + IMPROVED + PROMISE-STRENGTHENED HEADLINE]",
  "sub_headline": "[Supporting copy — 1 line max, reinforces headline]",
  "footer_text": "[Price • Offer • Guarantee — use exact site data]",
  "format": {
    "aspect_ratio": "1:1",
    "dimensions": "1080x1080",
    "platform": "Meta (Facebook/Instagram)",
    "ai_generator": "GPT Image 2"
  },
  "reference_image": {
    "provided": true,
    "instruction": "A reference image of the [PRODUCT NAME] is provided alongside this prompt. Recreate the [PRODUCT] PERFECTLY IDENTICAL to the reference image — same [KEY FEATURE 1], same [KEY FEATURE 2], same [FINISH/TEXTURE/MATERIAL]. Do not redesign, restyle, or alter any visual property of the product despite the surrounding scene."
  },
  "product": {
    "subject": {
      "description": "[FILL from Section 7 product JSON — exact product visual description]",
      "mirror_rules": "[FILL from Section 7 — what can/cannot appear in reflections]"
    },
    "accessories": {
      "prop": {
        "type": "[Product category]",
        "details": "[Precise visual details from Section 7]"
      }
    },
    "photography": {
      "camera_style": "[From Section 7]",
      "angle": "[From Section 7]",
      "shot_type": "[From Section 7]",
      "aspect_ratio": "1:1 square format",
      "texture": "[From Section 7]"
    },
    "background": {
      "setting": "[From Section 7]",
      "wall_color": "[From Section 7]",
      "elements": ["[From Section 7]"],
      "atmosphere": "[From Section 7]",
      "lighting": "[From Section 7]"
    }
  },
  "visual_format": "[Format name from Section 5]",
  "angle": "[1-3 word angle label — e.g. 'BOGO Urgency', 'Anti-Competitor', 'Desire / Third Party', 'Pain Elimination']",
  "consciousness_level": "L[1-5] — [Level name] — [1-sentence description of why this ad targets this level]",
  "belief_installation": {
    "primary_belief": "Belief #[N] — [Belief name] — [quote from brand research that embodies this belief]",
    "secondary_beliefs": [
      "Belief #[N] — [name] — [how this ad installs it]"
    ],
    "how_it_installs": "[Explain exactly how each creative element works to install the belief(s)]"
  },
  "scene_description": {
    "setting": "[Physical environment — room, background, props, surface]",
    "product_placement": "[Exact position, size, orientation, and visual treatment of the product in the scene]",
    "props": ["[Prop 1 with exact visual description]", "[Prop 2]", "[Prop 3]"],
    "environment": "[Lighting mood, color grading, depth of field, overall atmospheric description]"
  },
  "human_subjects": {
    "present": true,
    "count": 0,
    "description": "[Physical description matching brand's aesthetic and avatar aspirational vision]",
    "aesthetic_brief": "[Body type, style, age, ethnicity, wardrobe — all must align with avatar profile from Section 2]",
    "action": "[What they are doing in the scene — must be physically correct and contextually appropriate]"
  },
  "typography": {
    "headline_font": "[Font name + weight — e.g. 'Inter Black or Helvetica Neue Black']",
    "headline_size": "[Dominant — describe frame percentage and approximate px equivalent]",
    "headline_color": "[Exact hex + treatment — e.g. 'Pure white #FFFFFF with 2px black drop shadow at 30% opacity']",
    "headline_treatment": "[All caps / mixed / tracking / line height / position in frame]",
    "sub_headline_font": "[Font + weight]",
    "sub_headline_treatment": "[Size relative to headline, color, position, style]",
    "footer_font": "[Font + weight]",
    "footer_treatment": "[Size, color, separators, position]",
    "accent_text": "[Callout text specs — font, size, color, arrow style if applicable]"
  },
  "lighting": {
    "type": "[Lighting style — e.g. 'Cinematic editorial golden hour', 'Studio 3-point', 'Dramatic single-source']",
    "direction": "[Primary key light + secondary fill + any accent lights — direction, angle, intensity]",
    "mood": "[1-line atmospheric description]",
    "color_temperature": "[Kelvin values or color descriptors — warm/cool/neutral]"
  },
  "color_palette": {
    "primary": "[Hex]",
    "secondary": "[Hex]",
    "accent": "[Hex]",
    "background": "[Hex]",
    "rationale": "[2-3 sentences explaining the palette choice — psychological effect + alignment with brand + ad angle]"
  },
  "psychology_breakdown": {
    "scroll_stop_mechanism": "[Exactly what stops the avatar from scrolling in <0.3s — be specific about visual + cognitive + emotional triggers]",
    "emotional_trigger": "[Primary emotion activated + secondary emotion + how the ad exploits it for the avatar specifically]",
    "cognitive_hook": "[What question or tension does the headline create that demands resolution via clicking?]",
    "pattern_interrupt": "[How is this ad different from generic category ads that the avatar has already seen and ignored?]"
  },
  "viral_potential": {
    "score": 0,
    "rationale": "[Justify the score — what specific elements drive shareability, CTR, or retention? What risks reduce the score?]",
    "best_audience": "[Targeting brief — demographics, interests, platform, cold/warm/retargeting, geo]"
  },
  "funnel_stage": "[TOF / MOF / BOF — with 1-sentence explanation of why]",
  "differentiation_from_competitor": "[How does this ad differ from generic category/competitor ads? What makes it harder to replicate?]",
  "amazon_badges": {
    "include": false,
    "badge_type": "[Best Seller / Choice / none]",
    "placement": "[Exact position in frame, or 'N/A — [reason for exclusion]']"
  },
  "cta": {
    "primary": "[Explicit button text OR 'Implicit — clicking the ad']",
    "placement": "[Where the CTA appears in the ad]",
    "urgency_layer": "[Urgency element driving click — 'None' if not applicable]"
  },
  "guarantee_visibility": "[Where and how the guarantee appears in the ad]",
  "price_visibility": "[Is the price shown? Where? Why or why not?]",
  "press_logos": {
    "included": false,
    "list": ["[Press logo 1]", "[Press logo 2]", "[Press logo 3]"],
    "placement": "[Exact position and style in frame]"
  },
  "gpt_image_prompt": "[250-450 word paste-ready prompt following the structure from Section 8. Must include: image specs, reference image instruction, full scene, human subject if applicable, product placement, all typography with exact text, badges/logos, lighting, composition, mood, and the CONSTRAINTS block at the end (5-block structure from Section 8).]"
}
```


=================================================================
# SECTION 18 — TEMPLATE CLONE PROTOCOL (any product, any brand)
=================================================================

> Learned from 4 rounds of founder review (Oct 2026). Use this whenever the user gives template ads to re-adapt 1:1. It overrides the H1-H16 headline library: in clone mode, the template's own words are the starting point.

## STEP A — DECODE THE TEMPLATE (write it down before any prompt)
1. **Why does it work?** The scroll-stop trick in one sentence (visual joke, reframe, shame, desire, proof object, scarcity…).
2. **Template DNA — what belongs to THEIR product?** List every element of the image that exists only because of what they sell: the product itself, its material or texture (chocolate pour, colostrum milk, brisket slabs, serum drops, hair), its packaging, its brand logo, the body part of THEIR problem (arm, nose, hair, chin). **None of these may be imported.** Each one is swapped for OUR equivalent (our content, our proof object, our problem body part, our wordmark).
3. **What is the copy formula?** Rewrite the headline as slots: e.g. "[number] [clear rival product] a day just to [specific everyday task]. Then [us]." Every slot gets a concrete, specific value from the research: a rival the avatar really uses, a problem he really has, a routine he really has.
4. **Pick the angle that fits it best** from the research (never invent a promise or mechanism the research doesn't have).
5. **Humor / emotion check:** what does the image do against the text? Keep that exact mechanism (cheeky stays cheeky, shock stays shock).

## STEP B — ADAPT WITH MINIMUM CHANGE
- Change as little as possible: same words when the angle matches, swap only the nouns for the avatar's exact equivalents.
- Same amount of text as the template. No added context line, tagline or disclaimer.
- The image must SHOW the avatar's problem (belly, chest…), not the template's scene.
- Visualise the invisible: never a generic capsule — show what's inside (real ingredient pieces through a transparent capsule).
- Our content must look as appetizing / premium as theirs (no raw, ugly or ambiguous material; nothing that reads as "the animal was killed" or worse).
- Template has no product → add none. Template has a brand logo → our wordmark, spelled letter by letter, never a product photo in its place.
- The template's hero visual stays fully visible; the jar never covers it. Jar at real proportions, never squashed.

## STEP C — PROMPT STRUCTURE (GPT Image 2, images/edits)
Image 1 = product photo (lock). Image 2 = template. Then: TEMPLATE DNA line (their product + what not to import) → CHANGES (visual, incl. what NOT to copy) → TEXT slot by slot → CONSTRAINTS (only listed text, nothing from their product world, no extra jar, hero visible).

## STEP D — QC (full resolution, every image)
Product identical · text exact and same amount · image shows our problem · no element of their product world left · our avatar not their person · same render style · before/after difference obvious · hero visible · one-second test.

## STEP E — FEEDBACK LOOP
After each review: log what worked and WHY (to repeat it) and what failed + the root cause + the new rule. Show every version of a redone ad (B1 → B2 → …). Validated ads are not shown again.

## STEP 0 (before A) — ANTI-REGRESSION
- Read the full rulebook first: `/root/workspace/wh-remix/LESSONS.md` (also https://blueprint.machineroom.app/lessons). Every rule from every founder review stays active forever.
- Every ad spec MUST have the fields: why · dna · scene · visual · texts · engine (+ sexual: true when the template is suggestive).
- Run `python3 /root/workspace/wh-remix/lint.py <specs.py>` — any failure = nothing is generated until fixed.
- New review → add the rule to LESSONS.md FIRST, then to lint.py if machine-checkable, then generate.

## A3 bis — SCENE LOGIC
Ask why the scene is where it is (shower = hair in the drain; kid's room = product for kids' sleep). If the reason doesn't hold for our product, move to the scene where OUR problem shows (bathroom scale, closet, pool).

## A4 bis — SEXUAL / INNUENDO READING
Decode the hidden meaning (pineapple = vulva/taste, cream-cracked log = phallus, eggplant, "drain him"). Our version stays just as sexual AND ties to our angle. The visual must carry the same meaning as the text. Never sanitize it into something meaningless.

## D1 — ENGINE SWITCH
OpenAI refuses sexual images. Do NOT water the ad down: generate the scene with Flux 2 Pro via kie.ai (`kie_gen.py --model flux`, key /etc/secrets/kie_api_key), leave the product area empty, then composite the REAL jar cutout and exact small text with `composite.py`.
- Engine chain (B9): OpenAI → Nano Banana Pro (`kie_gen.py --model nano`, best at painting text INTO a scene) → Seedream (skin/lingerie scenes, render without text) → `composite.py` for the real jar + exact text.
- Lint now also blocks: invented deadlines, non-recognizable organs, missing `placement` field.