Todd Builds Kingdoms · AI use case · live demo
One real closet. Every piece got photographed on a couch, rendered clean, read for colour, scored in every combination and judged by an AI stylist, and now it's learning from what actually gets worn. Ask it for an outfit.
Pick an occasion or type it, add today's weather if you want, and you get four options back. Thumbs up, thumbs down, or say which one you're wearing, and it learns from that.
Each piece was shot on a couch with a phone and the engine re-drew it on a clean ground. Tap one to see what it's for and what it goes with.
Highest first. A look only scores 80 or above once a stylist or the owner has actually looked at it, so the rest carry an estimate and top out at 74.
Choosing "wearing this" logs the look, the occasion, the weather it used and the options you passed over, and pieces hit their wear limit and go to the laundry on their own. This demo keeps all of it in your browser only.
Built in about a week on top of Claude, a Postgres database, an image model for the renders and a free weather feed. Nobody hired a stylist; one got described.
Phone photos of each piece, wherever it lands, and a shoe rack photo is enough to identify a pair by model.
Each piece gets re-drawn as a clean catalog image and checked against the original so nothing gets invented.
Colour, pattern, fabric, how dressy it is, what occasions it can serve, how warm it is and whether it can take rain.
Every top with every bottom with every shoe, 7,104 combinations, each with a permanent code, scored on colour, structure, formality and brand rules.
An AI stylist grades the promising ones from the picture, 1 to 5, and says why, but the owner's own rating always wins.
Thumbs, what got worn over what got offered, and laundry state all feed back into every unseen look's estimate.
Codes are permanent, so L2334 stays L2334 through every re-score and an outfit can be named out loud. Location, when you use it, goes to the weather service only and isn't stored here.