json-image-prompting

name

json-image-prompting

description

Create structured JSON prompts for AI image generation, specialized for Hire Allstars marketing ads. Four modes: (1) Vision-to-JSON — analyze any competitor ad image with VisionStruct to produce a full visual blueprint, (2) Brand Transplant — take a competitor's VisionStruct JSON and swap in Hire Allstars brand colors, copy, and identity, (3) Idea-to-JSON — generate a Hire Allstars ad prompt from a brief or concept, (4) Generate — send the final JSON prompt to Nano Banana 2 (gemini-2.0-flash-preview-image-generation) and save the PNG. The full pipeline: competitor ad → VisionStruct JSON → Brand Transplant → Nano Banana 2 → Hire Allstars ad PNG. Triggers on: "analyze this ad", "replicate this", "make it Hire Allstars", "create ad for...", "generate ad", JSON prompt requests.

JSON Image Prompting — Hire Allstars Ad Engine

Turn any competitor ad into a superior Hire Allstars version using structured JSON prompts and Nano Banana 2.

Pipeline Overview

Competitor Ad (image/URL/screenshot)
        ↓  Mode 1: Vision-to-JSON
VisionStruct JSON (full visual blueprint)
        ↓  Mode 2: Brand Transplant
Hire Allstars JSON (brand applied)
        ↓  Mode 4: Generate
Nano Banana 2 → PNG saved to output/
        ↓
Review → Post to #open-claw

---

Mode Selection

Mode	Trigger	Input	Output
1. Vision-to-JSON	"analyze this ad" / image provided	Ad image	VisionStruct JSON
2. Brand Transplant	"make it Hire Allstars" / after Mode 1	VisionStruct JSON	HA-branded JSON prompt
3. Idea-to-JSON	"create ad for [concept]"	Text brief	Lightweight JSON prompt
4. Generate	"generate" / auto after Mode 2 or 3	JSON prompt	PNG via Nano Banana 2

Default flow when given a competitor image: Run Mode 1 → Mode 2 → Mode 4 automatically unless told otherwise.

---

Hire Allstars Brand Bible

Use these values in every Brand Transplant and Idea-to-JSON output.

Colors

Role	Hex	Usage
Primary Gold	#FCDC00	Headlines, CTAs, highlights
Deep Gold	#FCAF17	Gradient end, buttons
Dark Navy	#1a1a2e	Backgrounds, bold text
Pure White	#FFFFFF	Text on dark BG, card BG
Accent Blue	#4d65ff	Secondary CTA, links
Light Gray	#f8f9fa	Section backgrounds

Primary gradient: linear-gradient(135deg, #FCDC00 0%, #FCAF17 100%)

Typography

Font	Role	Weights
Lato	Body, captions, sub-headlines	300, 400, 700, 900
Bitter	Serif headlines, pull quotes	400, 700
Exo 2	Display headlines, CTAs	600, 700

Logo

Text-based wordmark: "HireAllStars"

• "Hire" in dark navy (#1a1a2e), bold
• "Allstars" with gold gradient fill (#FCDC00 → #FCAF17)
• No icon — wordmark only
• On dark backgrounds: reverse to white + gold

Voice & Messaging

Tagline: "Your Gateway to All-Star Talent at a Fraction of the Cost"

Core stats (always available for social proof):

• 80% lower staffing costs vs US equivalents
• 75% increase in hiring efficiency
• 2x higher retention rates
• 220+ headcount support teams built
• Featured: USA Today, Yahoo!, Newsmax, IBTimes

Value framing: Quality-first ("All-Star" talent), not budget-first ("cheap")

• Say: "Top-tier remote talent" / "All-Star candidates" / "Vetted professionals"
• Avoid: "cheaper", "budget", "affordable" (use "fraction of the cost" instead)

Primary CTAs: "Schedule a Call" · "Find Your Allstar" · "See Available Allstars" · "Book a Call"

Target audience for ads: US employers and hiring teams seeking remote talent in LATAM, Africa, and Asia.

---

Mode 1: Vision-to-JSON (VisionStruct Protocol)

Role

Act as VisionStruct — an advanced Computer Vision & Data Serialization Engine. Analyze the competitor ad with maximum fidelity. Capture every visual element as a machine-readable JSON blueprint for replication.

Core Directive

Do not summarize. Capture maximal visual data. Create a database record of the ad, not an art description. Every color, every text string, every spatial relationship.

Analysis Protocol (Silent Internal Processing)

1. Macro Sweep: Ad format, layout type, funnel stage (TOFU/MOFU/BOFU), primary message
2. Micro Sweep: Exact colors (hex estimates), font styles, text content (OCR), icons, decorative elements
3. Ad Sweep: Hook, offer, CTA, social proof elements, brand identity clues
4. Relationship Sweep: Spatial hierarchy, reading order, visual flow

VisionStruct Schema

{
  "meta": {
    "image_quality": "Low/Medium/High",
    "image_type": "Photo/Illustration/Typography/Mockup/Screenshot",
    "aspect_ratio": "1:1 / 4:5 / 9:16 / 16:9",
    "ad_format": "Static image / Carousel / Story",
    "funnel_stage": "TOFU / MOFU / BOFU",
    "ad_objective": "Awareness / Lead Gen / Conversion"
  },
  "global_context": {
    "scene_description": "Comprehensive paragraph describing the full ad layout and visual hierarchy",
    "layout_type": "Text-only / Hero image / Split / Overlay / Infographic / Screenshot",
    "background": {
      "color_hex": "#000000",
      "type": "Solid / Gradient / Photo / Pattern",
      "gradient_direction" 
    
     
  
   
       
     
     
     
  
   
     
     
     
     
  
   
     
       
       
       
       
       
    
     
       
       
       
    
      
     
       
       
       
       
    
     
     
  
   
    
       
       
       
       
       
       
         
         
      
        
    
  
   
     
     
     
      
     
  
   

---

Mode 2: Brand Transplant Protocol

Takes a VisionStruct JSON (Mode 1 output) and produces a Hire Allstars ad generation prompt.

Transplant Rules

Competitor Element	Hire Allstars Replacement
color_palette.dominant	Replace with HA brand colors (#FCDC00, #FCAF17, #1a1a2e, #FFFFFF)
text_ocr.headline	Rewrite with HA messaging, same format, sharper hook
text_ocr.subheadline	Rewrite referencing HA stats (80% cost savings, etc.)
text_ocr.cta_button	Replace with HA CTA ("Schedule a Call" / "Find Your Allstar")
text_ocr.social_proof	Replace with HA proof (80% lower costs, 2x retention, USA Today, etc.)
text_ocr.brand_name	Replace with "HireAllStars" wordmark
composition	PRESERVE — keep what's proven to work
layout_type	PRESERVE
objects[icons/decorations]	Match style but replace with relevant HA context

Output Format

The Brand Transplant outputs a ready-to-generate text prompt for Nano Banana 2, structured like this:

Generate a [aspect_ratio] image ad.

[Background description with HA brand colors]

The ONLY text on the image should be exactly:
- Headline: "[HA headline]" in [font/color/size]
- Subheadline: "[HA subheadline]" in [font/color]
- Body: "[HA body line 1]" / "[line 2]"
- CTA button: "[HA CTA]" — [button color] background, white text, [pill/rounded] style

[Visual elements description — icons, layout, decorative elements]

Style: [layout style from original — clean/infographic/etc]
Font: Lato or Exo 2 (sans-serif), bold headlines

NO competitor logos. NO competitor brand names. NO watermarks. NO extra elements beyond what is specified.

---

Mode 3: Idea-to-JSON

When given a brief or concept instead of a competitor image. Use the Hire Allstars Brand Bible to fill in all brand elements. Output the same generation prompt format as Mode 2.

Hire Allstars Ad Templates (Pre-Built)

Use these when the user asks for a specific format. Inspired by Somewhere.com's tested ad formats.

---

Template A — Bold Text Card (MOFU, employers)

Generate a square 1:1 image ad.

Dark navy background (#1a1a2e). Clean typography only — no photos, no icons, no borders.

The ONLY text on the image should be exactly:
- Line 1 (small, white, Lato 300): What US companies pay $80,000+ for...
- Line 2 (large, gold gradient #FCDC00→#FCAF17, Exo 2 700): You can hire for a fraction
- Line 3 (small, white, Lato 300): Vetted All-Star talent, ready to start in days.
- Line 4 (CTA pill, gold background #FCAF17, dark text, Lato 700): Schedule a Call

The "fraction" text should use the gold gradient. Center all text. Generous spacing between lines.

NO logos. NO brand names. NO watermarks. NO decorative elements.

---

Template B — Category Pills (TOFU, employers)

Generate a square 1:1 image ad.

White background. Clean, modern layout.

Top text (dark navy, Exo 2 600, medium): "We place All-Stars in:"
Below that, arrange pill-shaped tags in a grid (dark navy background, white text, Lato 400):
Sales · Finance · Marketing · Operations · Engineering · Admin · Customer Service · IT · Accounting

Bottom section (dark navy strip):
- Left: "HireAllStars" wordmark (white text, Lato 700)
- Right: Gold CTA pill "Schedule a Call"

NO photos. NO icons. NO extra copy. Pill tags should be evenly spaced.

---

Template C — Stats Mockup (MOFU, employers)

Generate a square 1:1 image ad.

Three-tier layout:
- Top band (dark navy #1a1a2e): Bold white text "All-Star talent at 80% less cost" + gold "80%" highlight
- Middle (white background): Clean mockup of a candidate profile card — name "Maria G.", role "Senior Operations Manager", skills "Process Optimization · Team Leadership · ERP Systems", rating 5 stars
- Bottom band (gold gradient #FCDC00→#FCAF17): Dark navy CTA text "Find Your Allstar →"

Style: Professional, product-forward, trust-building.

NO real photos. Illustrated profile card. NO extra text. NO watermarks.

---

Template D — Geometric Hero (TOFU/Awareness)

Generate a square 1:1 image ad.

Dark navy background (#1a1a2e). Gold partial-circle decorative shapes in top-right and bottom-left corners (using #FCDC00, semi-transparent).

Center-aligned text hierarchy:
- Line 1 (gold italic, Bitter 700): "Looking for elite remote talent?"
- Line 2 (white, Exo 2 300, smaller): "Work done right. Earn US-level results."
- Line 3 (white, Lato 400): "Join 500+ companies who hire All-Stars"
- Line 4: "HireAllStars" (white wordmark, Lato 700, with gold gradient on "Allstars")
- CTA pill (gold background, dark navy text, Lato 700): "Schedule a Call"

Premium, brand-forward aesthetic. Generous whitespace.

NO photos. NO extra elements. Clean geometric treatment only.

---

Template E — Infographic Flowchart (MOFU)

Generate a square 1:1 image ad.

White background with a clean infographic layout.

Serif headline at top (dark navy, Bitter 700): "Hire an All-Star in 3 Steps"

Three-step flowchart with icons and gold (#FCAF17) connecting arrows:
1. [Search icon] "Browse vetted candidates" — curated profiles, pre-screened
2. [Video icon] "Interview in days, not weeks" — fast, async or live
3. [Handshake icon] "Hire and onboard with support" — we handle the heavy lifting

Bottom: Stats row — "80% lower cost · 75% faster hiring · 2× retention"
CTA: Gold pill "Find Your Allstar"

NO logos. NO brand names in body. Clean flat icons, consistent line weight.

---

Mode 4: Generate (Nano Banana 2)

Model

Nano Banana 2 = gemini-2.0-flash-preview-image-generation

API Call

GEMINI_API_KEY=$(grep GEMINI_API_KEY .env | cut -d= -f2-)

curl -s "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash-preview-image-generation:generateContent?key=${GEMINI_API_KEY}" \
  -H "Content-Type: application/json" \
  -d "{
    \"contents\": [{
      \"parts\": [{
        \"text\": \"${PROMPT}\"
      }]
    }],
    \"generationConfig\": {
      \"responseModalities\": [\"image\", \"text\"]
    }
  }" \
  | python3 -c "
import sys, json, base64
data = json.load(sys.stdin)
parts = data['candidates'][0]['content']['parts']
for part in parts:
    if 'inlineData' in part:
        img = base64.b64decode(part['inlineData']['data'])
        open('${OUTPUT_PATH}', 'wb').write(img)
        print('Saved:', '${OUTPUT_PATH}')
        break
"

Output Path Convention

output/hire-allstars/ad-images/{NN}-{template-name}-v{N}.png

Examples:

• output/hire-allstars/ad-images/01-bold-text-card-v1.png
• output/hire-allstars/ad-images/02-category-pills-v1.png

Review Before Delivering

ALWAYS view the generated image before posting to Slack.

Checklist:

• All text is legible and correct (no garbled words)
• Colors match Hire Allstars brand (gold, navy, white)
• No hallucinated logos or competitor brand names
• CTA is visible and prominent
• Layout matches intended template
• No watermarks or extra elements

If issues: simplify the prompt, be more explicit about exclusions, regenerate.

---

Full Workflow (Competitor Ad → Hire Allstars PNG)

Step 1 — Receive competitor ad

• Image file, screenshot, or extracted from somewhere-com-complete-report.html

Step 2 — Vision-to-JSON (Mode 1)

• Analyze with VisionStruct schema
• Output: complete JSON blueprint

Step 3 — Brand Transplant (Mode 2)

• Apply Hire Allstars brand bible
• Rewrite all copy with HA messaging
• Output: Nano Banana 2 generation prompt

Step 4 — Generate (Mode 4)

• Run API call against gemini-2.0-flash-preview-image-generation
• Save PNG to output/hire-allstars/ad-images/

Step 5 — Review

• View PNG with Read tool
• Check against review checklist
• Iterate if needed (max 3 attempts per image)

Step 6 — Deliver

• Post to Slack: node bot/notify.js --file output/hire-allstars/ad-images/NN-name-v1.png --message "Ad NN — [Template Name] ready for review"
• Or package all ads: node bot/notify.js --file hire-allstars-campaign.zip --message "Hire Allstars campaign batch ready"

---

Competitor Ad Templates from somewhere-com-complete-report.html

Five employer-side Somewhere.com formats to replicate and improve:

#	Somewhere Format	Key Layout	HA Template to Use
1	Teal text-only ($500 headline)	Dark bg, bold text, gold accent	Template A
2	White minimalist typography	White bg, oversized type	Template A (light variant)
3	Infographic flowchart	White bg, arrow flow	Template E
4	Platform screenshot (role count)	3-tier card layout	Template C
5	Geometric abstract (teal + gold)	Decorative shapes, centered	Template D

---

Output Rules

1. VisionStruct JSON: Return in markdown code block, no preamble
2. Generation prompt: Return as plain text in a code block, ready to paste
3. After generating: confirm file saved, offer to post to Slack
4. For edits: show changed fields only (token efficient)
5. Always use output/hire-allstars/ as the root output folder