Kerbl, BernhardFranke, LinusHahlbohm, FlorianSteinberger, MarkusTagliasacchi, Andrea2026-04-222026-04-222026978-3-03868-267-71017-4656https://doi.org/10.2312/egt.20261003https://diglib.eg.org/handle/10.2312/egt20261003Creating high-fidelity digital twins from photographs has become increasingly practical thanks to recent advances in explicit 3D scene representations. Among these, 3D Gaussian Splatting (3DGS) stands out for its combination of high visual quality and extremely fast rendering. However, deploying such models in real applications reveals several challenges. A growing body of research addresses these limitations from multiple angles. Extensions to 3DGS improve portability, scalability, and stability; specialized techniques target flicker, popping, and aliasing; neural point-based methods introduce principled encodings that avoid several drawbacks of Gaussian primitives; and alternative explicit representations—–including triangles, stochastic splats, and novel volumetric structures—offer different trade-offs for varied use cases. This full-day tutorial provides a clear, consolidated overview of these fast explicit 3D representations. Attendees will learn how the major families of methods work, what constraints they address, and how to choose the right representation for specific interactive reconstruction or exploration tasks. Through conceptual explanations and end-to-end workflows, participants will gain practical guidance for using these techniques in their own research or pipelines, along with an understanding of their current limitations and open challenges.Attribution 4.0 International LicenseReal-time graphicsRadiance fieldsMachine learningRenderingParallel computing10.2312/egt.202610035 pages