Ad Ready

Ad-Ready: AI Advertising Image Generator

Generate professional advertising images from product URLs using a 4-phase AI pipeline on ComfyDeploy.

Source: github.com/PauldeLavallaz/ads_SV

Pipeline Architecture

The pipeline runs as a ComfyUI custom node deployed on ComfyDeploy. A single ProductToAds_Manual node executes 4 phases internally:

┌─────────────────────────────────────────────────────────────┐
│                  ProductToAds_Manual Node                     │
│                                                             │
│  PHASE 1: Product Scraping (Gemini Flash)                   │
│  ─────────────────────────────────────────                   │
│  Scrapes product URL → extracts title, description,         │
│  features, price, materials, image URLs                      │
│  Also scrapes HTML for high-res product images (≥1000px)    │
│                                                             │
│  PHASE 2: Campaign Brief Generation (Gemini Flash)          │
│  ────────────────────────────────────────────────            │
│  Brand Identity + Product Data + References →                │
│  10-point Campaign Brief (creative direction)                │
│                                                             │
│  PHASE 3: Blueprint Generation (Gemini Flash)               │
│  ──────────────────────────────────────────────              │
│  Master Prompt (funnel stage) + Brief + Keywords →           │
│  Production-Ready JSON Blueprint                             │
│                                                             │
│  PHASE 4: Image Generation (Nano Banana Pro / Imagen 3)     │
│  ──────────────────────────────────────────────────          │
│  Blueprint + all reference images → final ad image           │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Phase 2: Campaign Brief (The Creative Brain)

The Brief Generator is the most critical intermediate step. It acts as a "Senior Art Director" that translates raw data into actionable creative direction using a 10-point framework:

1. Strategic Objective — Why this campaign exists (awareness/positioning/launch)
2. Central Message — One idea perceivable without text
3. Visual Tone of Voice — Register: calm/energetic/intimate/monumental
4. Product Role — Hero vs co-protagonist vs implicit presence
5. Visual Language & Brand Coherence — Non-negotiable brand codes
6. Photographer & Equipment — Photography as concept, not execution
7. Extended Art Direction — Styling, casting, poses, hair/makeup, layout
8. Environment & Context — Where and why (conceptual, never decorative)
9. Texture, Material & Product Render — How surfaces are perceived
10. Final Image Signature — Finish, grain, temporal positioning

Without the brief, the Master Prompt must guess creative strategy. With it, the Master Prompt only executes.

The brief prompt template is included at {baseDir}/configs/Brief_Generator/brief_prompt.json.

Phase 3: Master Prompts (8 Funnel Stages)

Each funnel stage has a specialized Master Prompt that generates a production-ready JSON Blueprint. All share the same internal simulation:

• ROUND -1: Brand Identity Forensics (stages 03+) — Unified Brand Style Manifest
• ROUND 0: Fidelity Lock — Product geometry & talent identity are IMMUTABLE
• ROUND 1: Stage Strategy — Strategic approach specific to funnel position
• ROUND 2: Graphic Design — UI, typography, CTA engineering

The Blueprint JSON covers: scene production, talent lock, camera perspective, subject action/pose/wardrobe, lighting, product constraints, layout architecture, typography, CTA engineering, and brand asset placement.

Master prompt files are included at {baseDir}/configs/Product_to_Ads/.

Reference Analyzer

Reference images (referencia) are optional and off by default. The pipeline generates creative direction internally from Brand Identity + Campaign Brief. Only use a reference when the user explicitly asks to clone a specific ad's style.

When used, the reference is analyzed for pose, photographic style, and location cues.

⚠️ CRITICAL: Required Inputs Checklist

Before running ANY ad generation, ensure these are provided:

🚨 NEVER Skip These Steps:

1. Product image — Download the main product photo from the product URL. The scraper is fragile; always provide a product image explicitly.
2. Brand profile — If the brand doesn't exist in the catalog, run brand-analyzer skill FIRST to generate one. Never submit with "No Brand" when a brand is known.
3. Brand logo — TRY to find it (Clearbit, logo.dev, brand website). Use if good quality. If not found or low-res, skip it — the variable accepts empty string (bypassed server-side).
4. Reference — Do NOT search for references by default. Only provide when the user explicitly asks to clone a specific ad or says "find a good ad to clone".

Auto-Preparation Workflow

When the user asks to generate an ad:

1. User provides: product URL + brand name + objective

2. CHECK brand profile exists:
   → ls ~/clawd/ad-ready/configs/Brands/ | grep -i "{brand}"
   → If not found: run brand-analyzer skill first

3. DOWNLOAD product image:
   → Visit the product URL or fetch the page
   → Find and download the main product image
   → Save to /tmp/ad-ready-product.jpg

4. DOWNLOAD brand logo:
   → Search "{brand name} logo PNG" or fetch from brand website
   → Download clean logo image
   → Save to /tmp/ad-ready-logo.png

5. SELECT prompt profile based on objective:
   → 🎨 Morfeo_Creative: DEFAULT — cinematic, narrative-rich, slightly surreal. Best visuals.
   → Awareness: brand discovery, dynamic scenes, world-building, scroll-stoppers
   → Interest: sustained attention, micro-world hinting at use-case
   → Consideration: feature communication, proof cues, informative
   → Evaluation: trust, authority, reviews, certifications
   → Conversion: ⚠️ MINIMAL by design — clean, CTA-dominant, white backgrounds
   → Retention: post-purchase confidence, onboarding
   → Loyalty: editorial, lifestyle, emotional bond
   → Advocacy: share-worthy, community, belonging

   DEFAULT SELECTION LOGIC:
   - Generic "generate an ad" → Morfeo_Creative (09)
   - "awareness" / "brand discovery" → Awareness (01)
   - "conversion" / "buy now" / CTA-focused → Conversion (05)
   - "creative" / "original" / "surreal" → Morfeo_Creative (09)
   - "lifestyle" / "editorial" → Loyalty (07)
   - When in doubt → Morfeo_Creative (09), NOT Conversion

5b. SELECT MODEL (optional):
   → If user wants a person in the ad: pick from ~/clawd/models-catalog/catalog/images/model_XX.jpg (114 available)
   → If user wants product-only ad (no person): leave --model empty
   → If user doesn't specify: ASK if they want a model or product-only
   → Catalog preview: catalog.json at ~/clawd/models-catalog/catalog/catalog.json

6. RUN the generation with ALL inputs filled

Usage

Full command (recommended):

COMFY_DEPLOY_API_KEY="$KEY" uv run {baseDir}/scripts/generate.py \
  --product-url "https://shop.example.com/product" \
  --product-image "/tmp/product-photo.jpg" \
  --logo "/tmp/brand-logo.png" \
  --model "models-catalog/catalog/images/model_15.jpg" \
  --brand-profile "Nike" \
  --prompt-profile "Master_prompt_05_Conversion" \
  --aspect-ratio "4:5" \
  --output "ad-output.png"

With reference (only when explicitly requested):

COMFY_DEPLOY_API_KEY="$KEY" uv run {baseDir}/scripts/generate.py \
  --product-url "https://shop.example.com/product" \
  --product-image "/tmp/product-photo.jpg" \
  --reference "/tmp/reference-ad.jpg" \
  --brand-profile "Nike" \
  --prompt-profile "Master_prompt_01_Awareness" \
  --output "ad-output.png"

Auto-fetch mode (downloads product image and logo automatically):

COMFY_DEPLOY_API_KEY="$KEY" uv run {baseDir}/scripts/generate.py \
  --product-url "https://shop.example.com/product" \
  --brand-profile "Nike" \
  --prompt-profile "Master_prompt_05_Conversion" \
  --auto-fetch \
  --output "ad-output.png"

List available brands:

uv run {baseDir}/scripts/generate.py --list-brands

API Details

Endpoint: https://api.comfydeploy.com/api/run/deployment/queue Deployment ID: e37318e6-ef21-4aab-bc90-8fb29624cd15

ComfyDeploy Input Variables

Funnel Stages — Strategic Detail

01 — Awareness

Goal: Scroll-stop, curiosity, brand introduction Reject: Generic "product on table" concepts Strategy: Dynamic camera angles, world-building environments, high-concept creativity CTA: Soft or optional Visual Hierarchy: Talent → Product → Optional CTA

02 — Interest

Goal: Sustained attention, introduce value proposition Reject: Abstract visuals that hide the product Strategy: One clear visual idea, believable micro-world hinting at use-case CTA: Learn More, Discover, See Details Visual Hierarchy: Talent → Product → Headline → CTA

03 — Consideration

Goal: Informed evaluation, reduce uncertainty Reject: Pure mood storytelling, vague emotional content Strategy: Communicate WHAT product does, ONE primary differentiator, ONE proof cue CTA: Compare, See Details, Explore Visual Hierarchy: Talent → Product → Key Benefit → Proof Cue → CTA New: Adds Brand Identity Manifest to Blueprint JSON

04 — Evaluation

Goal: Validate purchase decision, proof & trust Reject: Pure mood, unsupportable claims, visual clutter Strategy: One trust anchor (quality/legitimacy/authority), one proof cue (reviews/certification) CTA: See Reviews, Verified Quality, Learn More Visual Hierarchy: Trust Anchor → Proof Cue → Product → Talent → CTA

05 — Conversion

Goal: Trigger decisive action, remove friction Reject: New hesitation-inducing info, complex compositions Strategy: One hero (product), one action, optional micro-reassurance CTA: Buy Now, Get Yours, Complete Order (PRIMARY visual element) Visual Hierarchy: Product → CTA → Optional Reassurance → Brand → Talent

06 — Retention

Goal: Post-purchase confidence, reduce churn Reject: Hard-sell, urgency, price talk Strategy: "You made the right choice" + "Here is the next step" CTA: Start, Set Up, Learn, Track (guidance, not purchase) Visual Hierarchy: Confirmation → Next Step → Product → Talent