A custom AI pipeline for a jewelry brand to generate product imagery from CAD files.
The brief
The brand designs jewelry in CAD before manufacturing. They wanted to test market response to new designs before committing to physical samples (which cost $400–800 each). They needed photoreal renders that customers couldn’t distinguish from real photography.
Our approach
We fine-tuned a custom image model on the brand’s existing real product photography (~3,000 photos). The model learned the brand’s specific lighting, surface treatment, and styling conventions. We integrated it with the brand’s CAD pipeline so a designer could go from CAD file to photoreal render in under 2 minutes.
The outcome
The brand tests new designs 6–10× faster. Sample manufacturing cost reduced by 40%. Three designs that previously would have been killed in CAD review went to market and became bestsellers.