The deep integration of generative artificial intelligence (AI) with publishing is driving a paradigm shift in the industry, fundamentally transforming its underlying logic and overarching forms. Against this backdrop, this paper introduces and systematically elaborates on the novel concept of "generative publishing". Generative publishing is defined as an emerging publishing paradigm that utilizes large language models as its technical foundation, is led by publishing institutions, adopts human?machine collaboration as its core content generation logic, and aims to produce generative publications and services tailored to users’ personalized needs. Its conceptual essence can be deconstructed across five dimensions: it relies on large language models as the technical base for autonomous content production; it redefines the role of publishing institutions from intermediaries to leaders in content creation, engaging directly with users; it prioritizes human agency within human?machine collaboration to ensure content quality and value alignment; its outputs manifest in two primary forms—generative publications and generative publishing services; and collectively, it signifies a distinct developmental phase beyond traditional publishing formats. On the basis of the dynamics of human?machine collaboration, the extension of generative publishing can be categorized into three types: (1) Controlled generative publishing: Human professionals retain full control, provide precise instructions and perform rigorous quality checks. The large language model functions as an advanced tool within a tightly defined human-led framework. (2) Interactive generative publishing: Characterized by a bidirectional feedback loop, this model involves iterative co-creation. The large language model generates content and performs initial assessments, while human experts provide selective calibration and creative input. (3) Autonomous Generative Publishing: The large language model operates with substantial independence within predefined objectives and ethical boundaries. Human oversight shifts from direct process control to ex-post review, exercising veto authority while relying on embedded governance rules. To operationalize this paradigm, this paper constructs an actionable framework comprising three core components: (1) The publishing large language model, serving as the technical cornerstone, developed through strategic planning, data preparation, training and fine-tuning, and deployment and maintenance. (2) Generative publications as the core products, including compiled types (intelligent reorganization and systematization of existing knowledge), derivative types (creative transformation of classical content), and original types (dual innovation in both conceptual and formal dimensions). (3) Generative publishing services, embodying the knowledge-service attribute, with representative forms such as knowledge service robots and intelligent reading interaction interfaces. Generative publishing represents the industry’ s evolution toward a dynamic, intelligent, and knowledge-service-oriented ecosystem. While promising, its advancement inevitably confronts challenges related to copyright, authenticity, professional restructuring, and ethical governance, necessitating a balanced approach that fosters innovation while upholding the core values of publishing.