DecoAR: An Intelligent Mobile Augmented Reality and Generative AI Framework for Spatial Interior Design

The prevailing consumer workflow for furniture selection and interior space planning still depends on manual dimension estimates and static visual judgment, which places a heavy cognitive burden on shoppers and contributes to elevated post-purchase return rates. This paper presents DecoAR, a native Android application that unifies markerless Augmented Reality (AR) with multi-modal Generative Artificial Intelligence (GenAI) to give users an in-situ, dimensionally accurate way to plan interior spaces. DecoAR is implemented with a declarative Jetpack Compose interface and is built on Google’s ARCore tracking subsystem, wrapped by the SceneView library and its Filament-based Physically Based Rendering (PBR) pipeline. The system contributes four coordinated capabilities: (i) markerless placement and gesture-based manipulation of 3D furniture assets on detected planes, (ii) a point-to-point spatial tape measure derived from 3D Euclidean geometry, (iii) real-time, flicker-free vertical-plane repainting for wall color testing, and (iv) an AI design assistant that reads the live Filament frame buffer and forwards it to Google’s Gemini 2.5 Flash model for context-aware layout feedback. Layouts are persisted locally through a reactive, GSON-serialized repository backed by shared preferences, so that users can save, reload, and share designs without a dedicated database engine. Empirical trials show sustained rendering at 60 frames per second (FPS) under standard furniture-placement workloads, a floor of 52 FPS under combined wall-painting and background AI transfer, and sub-2.1cm spatial measurement error even on low-texture surfaces. Unlike prior AR furniture visualizers, which function largely as passive object renderers, and prior generative-design tools, which reason over static images or floor-plan text rather than a live tracked scene, DecoAR couples geometry-grounded AR state directly to a vision-language model in a single closed loop. The results indicate that consumer-grade mobile hardware is now capable of hosting real-time spatial reasoning and generative design feedback within a single, responsive application.