In today's metaverse, where the digital and physical worlds blend seamlessly, capturing, representing, and analyzing 3D structures is vital. Advances in 3D and 4D tech have revolutionized gaming, AR, and VR, offering immersive experiences. 3D modeling bridges reality and virtuality, enabling realistic simulations and AR overlays. Adding time enhances experiences with lifelike animations and object tracking, shaping digital interactions.

Traditionally, 3D generation involved directly manipulating data, evolving alongside 2D techniques. Recent breakthroughs in 2D diffusion models have enhanced 3D tasks using large-scale image datasets. Methods like Score Distillation Sampling improve quality. However, biases in 2D data and limited 3D info pose challenges.

Generating 3D scenes and reducing biases in 2D data for realistic synthesis are ongoing challenges. Our tutorial explores techniques for diverse scenes and realism, including 3D/4D reconstruction from images and videos. Attendees learn about various generation methods, from 3D data training to leveraging 2D models, gaining a deep understanding of modern 3D modeling.

In summary, our tutorial covers the breadth of 3D/4D generation, from basics to the latest. By tackling scene-level complexities and using 2D data for realism, attendees gain insight into the evolving 3D modeling landscape in the metaverse.