A novel text-to-image generation method based on diffusion model is proposed to address the problems of low image fidelity,complex generation operation,and narrow applicability to specific task scenarios in existing text-to-image generation methods.This approach takes a diffusion model as the backbone network and designs a novel residual block structure to enhance generation performance.Additionally,a CBAM (Convolutional Block Attention Module) is integrated into the noise estimation network to improve the model's ability to extract key image information,thereby improving output quality.By combining conditional control networks,the approach achieves precise text-to-image generation with user-specific poses.Qualitative and quantitative analyses,along with ablation experiments,were conducted on the CelebA HQ dataset against methods such as KNN-Diffusion,CogView2,textStyleGAN,and Simple diffusion.Evaluation metrics and generation results demonstrate that,the proposed method effectively improves generation quality,with an average decrease of 36.4％ in FID (the Fréchet Inception Distance),average increases of 11.4％ in IS (Inception Score) and 3.9％ in SSIM (Structural Similarity).These results validate the effectiveness of the proposed approach.Furthermore,by integrating the ControlNet framework,the model enables text-to-image generation with controllable directional poses.