Problem
A European ecommerce company selling city-themed playmats at 45 euros each was trapped in a designer bottleneck. Their single designer spent eight hours creating each playmat in Illustrator, manually researching landmarks, verifying geography, and executing pixel-perfect illustrations. At that pace, they produced just 20 designs monthly, limiting catalog expansion and leaving potential revenue on the table.
Results
- 87% cost reduction: Design costs dropped from 256 euros to 34 euros per playmat
- 7x productivity multiplier: Same designer now produces 150 designs monthly instead of 20
- 11,000 euro monthly savings: At 50 designs per month, labor costs fell from 12,800 euros to 1,700 euros
- 3-minute AI execution: Full playmat generation in under 2 minutes, plus 20 seconds per regeneration
Before vs After
Before: Eight hours per design, 20 playmats per month, 256 euros per design, designer drowning in Illustrator execution work.
After: One hour per design (3 minutes AI + 1 hour polish), 150 playmats monthly capacity, 34 euros per design, designer elevated to creative director role.
Client Goal
The company wanted to expand their city playmat catalog without hiring additional designers. They needed a way to maintain design quality and geographic accuracy while dramatically accelerating production. The existing process was too slow to capitalize on market demand for new cities.
Challenges
- Geographic accuracy: City playmats require precise landmark placement based on real-world geography, not AI hallucinations
- Brand consistency: Every playmat needed to match an exact 11-color palette without random AI improvisation
- Designer burnout: Eight hours of manual execution work per design left no time for actual creative direction
- Catalog expansion limits: At 20 designs per month, the company couldn't keep up with customer requests for new cities
Solution Overview
We built a two-phase prompt engineering system that runs entirely in the Gemini interface. Phase one generates a verified geographic research brief with 23 landmarks. Phase two creates production-ready designs matching the 11-color brand palette. The designer went from eight hours of Illustrator execution to one hour of creative direction and quality control per playmat.
How It Works
- Phase One: Geographic Research Brief - Gemini receives a senior urban researcher role with geographic verification requirements. It researches terrain, road patterns, water features, and generates exactly 23 landmarks: 16 iconic monuments and 7 functional buildings. Every element must be verified on Google Maps to eliminate hallucinations. Output is a markdown design brief explaining why each landmark goes where based on real geography.
- Phase Two: Visual Generation - The research brief becomes input for phase two, which generates the actual playmat using the hardcoded 11-color palette. The prompt breaks down every standardized element: roads, roundabouts, grass, and water. Gemini follows the rulebook instead of improvising.
- Designer Quality Control - If errors appear, the designer clicks regenerate for a new version in 20 seconds. Average is two regenerations, so three minutes total for AI execution. Then one hour of polish using Nano Banana for final adjustments.
- Catalog Expansion - Same process applies to infinite city variants. The research phase changes per city, the visual generation phase stays the same. 80% acceptance rate means most designs work with minimal iteration.
Key Features
- Constraint engineering over randomness: 11-color palette and geographic verification eliminate infinite option paralysis
- Phase separation prevents context collapse: Research and visual generation are distinct cognitive tasks with full context windows
- Designer role elevated: From eight hours of execution to one hour of creative direction
- Template scalability: Once built for one theme, it applies to race circuits, stadiums, theme parks with minimal modification
- No workflow tool required: Everything lives in Gemini interface, just two copy-paste prompts and a human checkpoint
Tools Used
Gemini
AI model executing both research and visual generation phases with constraint engineering to eliminate hallucinations
Google Maps
Verification layer ensuring every landmark is geographically accurate and actually exists in the real city
Nano Banana
AI-powered image editor for final adjustments and polish, reducing manual correction time from hours to minutes
Video Walkthrough
Ready to Multiply Your Design Output?
If your designer is drowning in execution work instead of creative direction, we'll show you how to engineer the constraints. Turn one designer into a design factory with AI prompt systems.