07/21/2026
Every project teaches us something. Here's what building ICON Golf Cars' image pipeline taught us about where AI tools actually earn their keep.
ICON's new i40FX was still in production, so the only imagery available was the standard manufacturer product renders — one shot per color, clean white background. Normal for that stage. Not something to build a launch campaign on.
Here's the old way to fix that: take a flat product shot and composite it into a real-looking scene — correct reflections, correct shadows, a background that actually belongs there. A skilled retoucher can do that. One image, done well, is real time and real money. Do it for every color, every dealership market, and you're looking at weeks of manual compositing work, if you can find the hours to give it.
We took a different path. Used Claude, Grok, and Gemini to write and refine the scene descriptions — what a Forsyth Park morning in Savannah actually looks like, what an Aberdeen, NC neighborhood street actually looks like. Tested that against four different image generation tools — Higgsfield, Google Veo, Gemini, and Grok — until we found the combination that held up: correct light direction, real reflections in the windshield glass, backgrounds that matched the actual market. Once we had it, we turned it into a repeatable process — what we call a prompt bible — so it wasn't a one-off, it was a system.
Then we ran it. Hundreds of location-specific images, each one built for its market, in a matter of hours and days — not weeks.
Hundreds of images meant hundreds of approval decisions, so we built a review portal ourselves — no third-party tool, just a link where the team could approve or pass on each image and every decision got tracked automatically.
This is what we mean when we say AI tools plus a real system beats either one alone.
Sitting on product assets that feel thin and need to look market-ready fast? Send us a message — we'll show you exactly how the pipeline would apply to your product line.