Advancements in 3D World Generation

The field of 3D world generation is rapidly advancing, with a focus on creating detailed, realistic, and scalable models. Recent developments have centered around the use of multimodal agent frameworks, which leverage diverse foundation tools to acquire real-world knowledge and construct complex 3D scenes. These frameworks have shown significant improvements in reality alignment, shape precision, texture fidelity, and aesthetics. Noteworthy papers include RAISECity, which achieves a 90% win-rate against existing baselines for overall perceptual quality, and MajutsuCity, which reduces layout FID by 83.7% compared to CityDreamer. Yo'City is also notable for its ability to generate personalized and boundless city-scale scenes through a top-down planning strategy and user-interactive expansion mechanism. FilmSceneDesigner has also made significant contributions by automating film set design through an agent-based chaining framework. Overall, these advancements have the potential to revolutionize applications in immersive media, embodied intelligence, and world models.

Sources

RAISECity: A Multimodal Agent Framework for Reality-Aligned 3D World Generation at City-Scale

Yo'City: Personalized and Boundless 3D Realistic City Scene Generation via Self-Critic Expansion

FilmSceneDesigner: Chaining Set Design for Procedural Film Scene Generation

MajutsuCity: Language-driven Aesthetic-adaptive City Generation with Controllable 3D Assets and Layouts

Built with on top of