The Challenge
The brand needed massive amounts of custom 3D imagery to feed a high-speed digital publishing schedule. Traditional rendering took days. Off-the-shelf AI could not replicate our specific brand style. We had strict 24-hour turnaround times and a 3D design team on the verge of burnout. We needed scale without sacrificing quality.
The Build
To solve our challenge I engineered a custom Adobe Firefly model using a recursive training loop to teach the AI our exact visual language.
Phase 1
The Foundation
Trained a baseline Firefly model on 10 core texture renders from our brand 3D library.
Phase 2
The Scaffolding
Generated basic 3D shapes in Illustrator. This forced the AI to learn how to wrap our proprietary texture around actual structures.
Phase 3
The Recursive Loop
Used Gemini to analyze the outputs, optimize the prompts, and write advanced training captions. I fed the successful AI-generated shapes right back into the model as new training data. By escalating from simple spheres to complex objects, the model learned to generate highly detailed structures built entirely out of our brand texture.
The Tool
I did not want this to live in a silo. I collaborated with Gemini to synthesize the optimized prompts into a master formula, then built a custom HTML “Mad Libs” interface. Suddenly, any designer could generate consistent brand 3D assets without touching a rendering program.
The Impact
Unlocked Speed: Cut custom 3D generation from a multi-day render to a sub-24-hour workflow.
Total Access: Empowered 2D designers to build complex, brand-compliant 3D objects with zero prior modeling experience.
Protected Sanity: Completely eliminated the 3D bottleneck. The core motion team went back to high-craft campaign work instead of drowning in high-volume social requests.
This is the personal site of Nick Burns and is in no way affiliated with or endorsed by Red Hat, Inc.
