With the rapid development of 3D display technology, there is an escalating demand for high-quality 3D object generation technology. Traditional manual modeling techniques, reliant solely on skilled artists, are inadequate to cater to the burgeoning requirement for diverse 3D assets. Consequently, there is an urgent need for convenient and cost-effective approaches to intelligent 3D object generation. This paper proposes a new text-driven mesh autogeneration model for generating 3D objects with textures. Leveraging deep learning architecture, our model facilitates the end-to-end generation of both geometries and surface materials of 3D objects in response to textual prompts. Existing text-to-3D generation models lack the ability to decouple materials, for which we investigate the effects of network frequencies and vertex offsets on the generation results and propose a control factor \({\sigma }\) for textures and a control factor \({\beta }\) for geometric details, which enable accurate generation of objects with different materials. Experimental evaluations validate the model’s capability in geometric generation and texture synthesis, demonstrating cutting-edge performance in mesh quality and texture fidelity.