OrbitForge: Text-to-3D Scene Generation via Reconstruction-Anchored Video Synthesis

OrbitForge is a novel adapter that leverages frozen video priors and Gaussian Splatting reconstruction to generate 3D scenes from text inputs, addressing the limitations of traditional text-to-video models. By harnessing the power of generic text-to-video models as open-world scene priors, OrbitForge enables more accurate and consistent 3D asset creation. The adapter's ability to control camera motion and provide comprehensive view coverage mitigates the inconsistencies often found in frames generated by traditional models. This breakthrough has significant implications for various applications, including computer-aided design, video game development, and simulation training. The introduction of OrbitForge marks a substantial advancement in the field of text-to-3D scene generation, offering a more reliable and efficient means of creating complex 3D environments¹. This matters to practitioners as it enables the creation of more realistic and immersive virtual worlds, which can be used to enhance training simulations, architectural visualizations, and other applications that rely on accurate 3D modeling.