Smartphone-Based 3D Gaussian Splatting of a Monumental Royal Interior: A Frame-Density Study at Palazzo Reale di Torino

High-quality 3D documentation of monumental heritage interiors typically requires expensive professional equipment. This paper proposes a budget-accessible 3D Gaussian Splatting pipeline relying solely on a consumer smartphone and off-theshelf laptop GPUs, demonstrated on the Appartamento del Re at the UNESCO-listed Palazzo Reale di Torino. A single 15-minute walkthrough video was processed through eight uniformly sampled frame subsets (n from 450 to 3 000). Quality improves monotonically with density: n ≥ 900 frames are required for a connected pose graph, n ≥ 1 800 for high-fidelity surfaces, and n = 2 100 best matches a Matterport Pro3 reference floor plan, reaching PSNR = 28.2 dB and SSIM = 0.942 on 179 held-out views. As all subsets saturate the 3M primitive cap, quality is governed by camera-pose accuracy rather than primitive count; the stochastic disconnection of entire rooms from the pose graph remains the main reliability challenge. The reconstruction renders in Unreal Engine 5 at 26.5 fps on an entry-level laptop GPU, limited by CPU-side depth sorting.