LIVE DEMO · AI × WHITE-LABEL
Type one sentence about an imaginary brand. An AI model art-directs it as a config: palette, type, copy, catalog, even image prompts. The storefront below obeys the config and nothing else. This is my white-label architecture with an AI hand on the config file.
Try: a luxury watch house from Geneva · a neon arcade bar for retro gamers · an organic farm box for busy parents
The storefront never changes. Only the config does. That separation is what let one real codebase run nine production brands: when brand is data, adding a brand is writing data, and an AI can write data. The interesting work is the architecture that makes the config the only thing that needs to change.
Images are the one thing a config cannot contain, so the config carries prompts instead. The storefront draws procedural art from the palette the instant the config lands, then fades in photos generated from those prompts as they arrive. If an image never comes, nothing breaks.
The plumbing here is the same shape I build for clients: the model runs server-side behind an API route, the key never reaches the browser, output is schema-locked and validated (including a contrast check that corrects any palette the model gets wrong), and a rule-based generator takes over if the API misbehaves.
See the pattern at production scale: One build, every brand →