SceneWiz3D is a novel approach that synthesizes high-fidelity 3D scenes from text descriptions. It employs a hybrid 3D representation, using explicit representations for objects and implicit representations for scenes. Users can generate objects through conventional text-to-3D methods or by providing their own models. To configure scene layout and automatically place objects, we leverage particle swarm optimization during the optimization process. Furthermore, in text-to-scene scenarios, certain parts of the scene (e.g., corners, occluded areas) often lack multi-view supervision, leading to inferior geometric quality. To mitigate this supervision deficiency, we introduce an RGBD panoramic diffusion model as an additional prior, enabling the generation of high-quality geometry. Extensive evaluations demonstrate that our method achieves higher quality compared to previous approaches, producing detailed and view-consistent 3D scenes.