ViewCrafter

ViewCrafter is a novel method for synthesizing high-fidelity novel views of generic scenes from single or sparse images using a video diffusion model prior. It combines the generation capabilities of video diffusion models with coarse 3D clues from point-based representation to produce high-quality video frames with precise camera pose control. The method includes an iterative view synthesis strategy and a camera trajectory planning algorithm to progressively extend 3D clues and covered areas. Applications include immersive experiences with real-time rendering via 3D-GS optimization and scene-level text-to-3D generation. Extensive experiments demonstrate strong generalization and superior performance in synthesizing consistent novel views.