Champion is one of America's most iconic sportswear brands. At their scale, creative production means hundreds of SKUs, multiple colorways, multiple silhouettes, multiple shot types, all of it needing to be consistent, accurate, and ready to use.
Company
Champion
Timeline
2023
—
2025
Role
AI Pipeline Architect · Technical Creative Lead · Systems Design
Project overview
I built the AI production pipeline for Champion's garment recoloring system from the ground up, and that started before I wrote a single line of code. First I wrote a PRD for myself, mapping out the full logic of the system, what inputs it needed, what outputs it had to produce, how errors should be handled, and what the QA layer needed to catch. Then I designed the data schema and job architecture, defining how each garment's metadata, colorway targets, shot type, and prompt variables would be structured and passed through the system. Only after that did I build.
What started as browser-based JavaScript injection evolved into a Python runner connected to a JSON jobs file and the ComfyUI API, a system that could process hundreds of product images with hex-accurate garment recoloring, tone-on-tone logo handling, and shot-type-specific prompting across full, front, back, and detail views. Per-image audit data handled automatically. The whole thing built to run without someone babysitting every step. I didn't just generate images. I designed the logic, mapped the data, and built the infrastructure that generates images reliably at scale.
Challenges
Color accuracy at this volume is unforgiving. A garment that renders at the wrong hex across 200 images isn't a creative miss, it's a production failure. The schema design had to account for edge cases before they happened, tone-on-tone logos, shot types that behave differently under the same prompt, colorways that read differently on different fabric textures. Getting that right meant thinking through the failure modes before building the system, not after.
Results
Hundreds of production-ready recolored assets delivered. The pipeline architecture and data schema are now the reusable foundation for production work.
Latest projects
Champion is one of America's most iconic sportswear brands. At their scale, creative production means hundreds of SKUs, multiple colorways, multiple silhouettes, multiple shot types, all of it needing to be consistent, accurate, and ready to use.
Company
Champion
Timeline
2023
—
2025
Role
AI Pipeline Architect · Technical Creative Lead · Systems Design
Project overview
I built the AI production pipeline for Champion's garment recoloring system from the ground up, and that started before I wrote a single line of code. First I wrote a PRD for myself, mapping out the full logic of the system, what inputs it needed, what outputs it had to produce, how errors should be handled, and what the QA layer needed to catch. Then I designed the data schema and job architecture, defining how each garment's metadata, colorway targets, shot type, and prompt variables would be structured and passed through the system. Only after that did I build.
What started as browser-based JavaScript injection evolved into a Python runner connected to a JSON jobs file and the ComfyUI API, a system that could process hundreds of product images with hex-accurate garment recoloring, tone-on-tone logo handling, and shot-type-specific prompting across full, front, back, and detail views. Per-image audit data handled automatically. The whole thing built to run without someone babysitting every step. I didn't just generate images. I designed the logic, mapped the data, and built the infrastructure that generates images reliably at scale.
Challenges
Color accuracy at this volume is unforgiving. A garment that renders at the wrong hex across 200 images isn't a creative miss, it's a production failure. The schema design had to account for edge cases before they happened, tone-on-tone logos, shot types that behave differently under the same prompt, colorways that read differently on different fabric textures. Getting that right meant thinking through the failure modes before building the system, not after.
Results
Hundreds of production-ready recolored assets delivered. The pipeline architecture and data schema are now the reusable foundation for production work.