The Challenge
A shopper browsing a marketplace one item at a time is doing more work than a shopper should have to do to picture a finished look, and the company saw an opportunity that went further than fashion discovery alone: build visual AI capable enough to also power image generation, enhancement, ranking, and quality control across the rest of the platform, so the investment in one feature could pay for capability the whole business could use.
The Approach
Solvd helped build the production visual-commerce system behind the company’s outfit-discovery feature and the related capabilities that followed it. The system generates complete outfit imagery at scale using custom-trained models, links every generated look back to real inventory so a shopper can actually buy what they’re looking at, and runs automated checks for hallucinations, product mismatch, unsafe content, and invalid inputs before anything reaches a buyer. The same underlying platform grew to support image enhancement, background editing, aesthetic ranking, personalization, and browse optimization elsewhere on the site, which is exactly the leverage the company had been hoping the initial investment would create.
The Result
Generated outfit imagery increased add-to-cart activity, and the visual ranking and image-optimization work built on the same foundation improved conversion and engagement across search, browse, SEO, and marketing, which is the kind of compounding return a platform investment is supposed to produce and rarely does this cleanly.
Automated quality control kept defective generated images below 5 percent without manual review, at the volume a marketplace this size actually needs.
