Driven by the national science popularization strategy and the in-depth digital–intelligent transformation of the publishing industry, traditional printed science popularization books confront challenges of single-dimensional dissemination and weak interactivity, thereby prompting a paradigm shift from static knowledge storage toward dynamic, user-oriented communication. This paper explores how generative AI can be harnessed to enable the secondary creation and multimodal transformation of science popularization content, aiming to address the practical dilemmas of limited coverage and high cognitive thresholds in traditional science popularization. This study first constructs a systematic practical framework for AI-enabled content transformation, organized around four interlocking dimensions: content structuring, media adaptation, technological implementation, and feedback iteration. Its core logic lies in transforming the linear, systematized knowledge embedded in printed books into decomposable and recombinable knowledge fragments. Through content deconstruction, science popularization content can better adapt to diverse communication media and achieve the reconstruction of its expressive forms. The research indicates that the core of this transformation lies in upholding professional accountability and advancing human–AI collaborative creation. With the support of a data-driven closed-loop user feedback mechanism, publishing institutions can achieve continuous iteration and optimization of science popularization products. Furthermore, this study categorizes the specific forms of AI-powered secondary creation into three types: short-video adaptation, audio adaptation, and AI interactive assistants. Specifically, short videos visualize obscure or complex knowledge points by leveraging AI-generated video scripts and visual materials; audio adaptation converts textual content into accessible audio-based science popularization products; and intelligent interactive assistants upgrade static knowledge into an immersive, inquiry-driven learning experience. Together, these three forms substantially lower the barriers to understanding scientific concepts and effectively boost user engagement, aligning closely with the reading and learning habits of diverse audience groups. Meanwhile, this paper analyzes the inherent risks of integrated AI application, including AI hallucinations, potential copyright infringement, and the weakening of human editorial subjectivity. To mitigate these risks, three targeted countermeasures are proposed: first, strengthening professional editorial review and supervision; second, standardizing the compliant management of data authorization; and third, promoting the shift of AI application from individual trials to large-scale institutional practice. Such management ensures the rigor of scientific content while improving efficiency through technology. Accordingly, AI-driven multimodal transformation functions as a crucial carrier for realizing the digital–intelligent inclusion of scientific knowledge, enabling people from all walks of life to access high-quality science popularization resources. This paper argues that publishing institutions should redefine the value of artificial intelligence as an industry enabler rather than a human replacement. By restructuring business workflows and reasserting the subjectivity of publishers, the industry can strike a balance between technological efficiency and professional ethics. The research framework advanced in this paper provides feasible solutions for adapting to the evolving media environment and offers a forward-looking exploration of reconstructing the popular-science communication paradigm in the digital–intelligent era.