The rapid spread of generative artificial intelligence is reshaping how content is produced and disseminated on social media platforms, intensifying challenges such as cross-platform cascading diffusion of misinformation, rising moderation costs, and blurred accountability chains. Current practice largely relies on linear “human + algorithm” governance and minimal compliance. Addressing gaps in existing research—namely the lack of mapping between technical architectures and governance tiers, overly static collaboration mechanisms, and the difficulty of parameterizing and transparently assessing the safety-efficiency-fairness trade-offs—this paper proposes an integrated, four-tier collaborative governance framework based on the co-construction of “technical architecture-governance level.” At its core, the framework establishes a clear correspondence of “who is responsible—what is governed—how it is governed—how it is evaluated,” linking technical layers (data, models, content distribution) with governance tiers (institutional/regulatory, industry, platform, societal). It is supplemented by a bidirectional governance loop (top-down rules and bottom-up risk signals) and cross-platform coordination mechanisms, translating abstract governance principles and values into measurable, adjustable, and auditable metrics and weighting schemes. This enables multi-stakeholder collaboration to be operationalized as executable and assessable meso-level designs, bridging gaps between regulations, standards, and platform operations; supporting agile transitions between routine and crisis contexts; shortening response latency; curbing cascading spread; and improving exposure fairness and external accountability at a manageable efficiency cost. The framework provides a theoretical foundation and a systematic analytical toolkit for platform governance in the generative AI era.