In recent years, image generation technology has developed rapidly. Using image generation technology for style transfer and image synthesis can effectively improve the creative efficiency and innovation ability in fields such as advertising design and film and television production. However, in terms of protecting traditional culture such as embroidery skills, existing image generation methods mainly rely on style transfer techniques, which are difficult to effectively restore real details, and have limited generation capabilities. To address this issue, this paper proposes an embroidery image generation method based on texture and text conditions. This method introduces a diffusion model for the first time to restore embroidery pattern details and designs an efficient diffusion model structure. By using fine-grained dual channel feature fusion technology, the embroidery texture and text description are truly restored. This method not only improves the ability to reproduce details in generated images, but also reduces creation costs and technical barriers, promoting innovation and development of traditional culture such as embroidery in the field of digital applications.