LucidDreamer is a domain-free 3D scene generation technique that leverages the capabilities of existing large-scale diffusion generation models. It can generate navigable 3D scenes from a single text prompt or a single image. The method employs a dream-alignment two-step procedure, first generating multi-view consistent images based on the input, and then harmoniously integrating the newly generated 3D scene parts together. LucidDreamer generates highly detailed Gaussian splat maps, without the limitations of target scene domains compared to previous 3D scene generation methods.