Pinterest launches Restyle, an AI room-design feature that tries to close the pin-to-purchase loop

The engine behind the scenes runs on what Pinterest calls Pinterest Intelligence, infrastructure built with Nvidia on Blackwell processors, Dynamo software, open models and home-grown tech. The company says the stack sharpens its visual assistant's ability to handle visual search and turns user queries into AI-driven discovery and purchase signals. Pinterest announced the infrastructure on Monday, three days before the consumer product dropped.

Here is how it works. Users snap a photo of their room, then tell the model to add art, furniture, accessories or plants — including items they have already saved on Pinterest — or to change lighting, wall color and overall style: "bohemian," "industrial," "whimsical" or "dopamine." They can also tap a single object in the image to swap, delete or edit it in isolation instead of regenerating the whole scene. The approach feels more like layer editing than flat image generation.

AI room design is already a common use case in other assistants. Pinterest's edge is the save repository it already owns: millions of users have curated inspiration boards, and now they can "dress" those saved items onto their actual space. The company bets the home visualization will lift real-world purchase odds and finish the journey from pinning to checkout, a loop the platform has been tightening for a while.

Restyle ships in early beta in the U.S. and Canada, with a broader rollout slated for next month. The reveal came at the annual Pinterest Presents event, where the company also debuted new visual-search ads, app-promotion tools and additional advertiser features — a signal that the consumer product and the ad infrastructure are built to work together.