Abstract: The task of text-guided customized 3D model generation for specific-subjects poses a fundamental chal-lenge: existing methods struggle to balance identity consistency with viewpoint diversity. To address this, a personalized text-to-3D model generation method is proposed for specific subjects. First, a diffusion model is fine-tuned on images of the target subject to learn its identity features, yielding a subject-specific per-sonalized image generation model; second, a large language model analyzes the input text description, and a Gaussian-based initialization module is employed to generate a clean front-view image with a prominent subject; third, layout control information is injected into the cross-attention layers of the diffusion model to constrain the position and completeness of the subject in the generated image; finally, an image-to-3D modeling technique is applied to predict multi-view information from the front-view image and reconstruct the final 3D model. Experiments on the DreamBooth dataset demonstrate that the proposed method achieves a CLIP-I score of 0.697 and a CLIP-T score of 0.295, significantly outperforming existing meth-ods including DreamBooth3D, MVDream, and GaussianEditor in terms of both geometry accuracy and visual fidelity, validating the effectiveness of the proposed approach.