Overview
Flur is an AI-assisted 3D editor for architecture and interior design. A user uploads a floorplan as a PNG, JPG, or DWG file, and a Google Gemini-driven pipeline reads the drawing and reconstructs the space as a live, editable 3D scene. It is built for architects, interior designers, and project teams who need to move from a flat plan to a navigable model without rebuilding it by hand.
The Problem
Turning a 2D floorplan into a usable 3D model normally means manually rebuilding walls, openings, slabs, and furniture in specialized CAD or modeling software. Flur automates that reconstruction so an uploaded drawing becomes an editable, photorealistic scene in a single step.
Key Features
- Floorplan reconstruction — a Gemini-driven analysis pipeline reads uploaded PNG, JPG, or DWG floorplans and a node-based building engine rebuilds walls, openings, slabs, and furniture as a live scene graph.
- Real-time 3D viewer — the reconstructed model renders in a WebGPU and Three.js (React Three Fiber) viewer that stays fully editable as the underlying scene graph changes.
- Photorealistic renders — GPU path tracing produces photorealistic beauty renders directly from the working scene, without exporting to a separate rendering tool.
- Immersive walkthroughs — Gaussian-splat room tours and 360-degree panorama walkthroughs let clients move through a designed space interactively.
- Furniture and brand catalog — a built-in furniture and brand catalog lets users place cataloged products directly into the scene as they design.
- Versioned history and i18n — versioned design history tracks changes across a project, and the interface is fully localized in English and Chinese.
How It's Built
Flur is a pnpm and Turborepo monorepo built on Next.js, React, and TypeScript, with Tailwind CSS for the interface and Supabase for data and storage. The 3D layer runs on Three.js and React Three Fiber over WebGPU, adding GPU path tracing for photorealistic output and Gaussian splatting for immersive room tours. A Google Gemini pipeline drives floorplan analysis, feeding a node-based building engine that assembles walls, openings, slabs, and furniture into an editable scene graph.
Tech Stack
Note
Built for Horizon Analytics Labs; the source is private, so this write-up stays at a high level.