In recent years, text-to-3D generation has made great progress and can generate many exquisite 3D objects. However, due to the weakness of text encoder in processing long text, the task of text-to-3D generation with complex attributes has always encountered great difficulties. In particular, when there are serious occlusion relationships between these complex attributes, the results will be worse. Therefore, we propose a new method called Hierarchical-Chain-of-Generation (HCoG), which needs no manual efforts, utilizing large language model to decompose complex target objects into a hierarchical generation chain, so that each part can be better generated. Furthermore, for each split text, SAM automatically find the corresponding region and optimize 3D Gaussian kernels in this region by a controllable way. In addition, to generate new parts in the hierarchical chain, we need to preserve previous parts and optimize new parts. Therefore, we propose the Label Elimination to ensure new parts will not attach to the surface of the previous parts and change them. Experiment demonstrates that HCoG is an end-to-end automatic framework for advanced complex attributes text-to-3D generation while effectively handling situations where there are a lot of occlusions between attributes and ensuring high quality of the results.