In response to the efficiency and creativity limitations faced in traditional art creation, as well as the subjectivity of manual art creation, the GAN (Generative Adversarial Network) generative model driven by AI (Artificial Intelligence) is introduced for image generation. The purpose is to convert text descriptions into images through machine learning models, break through the limitations of traditional creation, and promote new possibilities for interdisciplinary art creation. First, this paper builds a system by integrating NLP (Natural Language Processing) and GAN technology, uses natural language processing technology to process text descriptions, and uses a pre-trained language model to extract semantic features in the text to ensure the semantic alignment of text and image. Then, using the generative adversarial network model, the processed text information is input into the generator, and after feedback from the discriminator, the quality and diversity of image generation are gradually optimized. In this process, the conditional generation technology (Conditional GAN) is also introduced to ensure the diversity and creativity of the generated images through the control of the conditional vector. Finally, the quality of the generated images is scored and optimized through human evaluation and automatic assessment methods to improve the artistic expression and innovation of the generated system. Experimental results show that the automatic generation system based on text to image can effectively generate image works with high creativity and artistic sense. Most of the generated images can match the text description well. For example, “The beach under the sunset” scored 4.8 points, indicating that the image successfully captured the light and scene in the text. In summary, AI-driven art creation systems, especially text-to-image generation technologies, not only have great potential in artistic creation, but also provide new tools and sources of inspiration for artists, designers and creative industries.