How our conscious experience materialize from our predictions is an open question in neuroscience and philosophy. Here we test the hypothesis that top-down predictions arise from a trio of domain-specific generative models (Fig 1a) within prefrontal (PFC) nodes of the Default-Mode Network (DMN), and integrate within its parietal nodes, forming the basis for our dynamic consciousness.Smoothed binary ratings of prediction updates for State changes, Agent beliefs, and Action sequences from an online sample were used as parametric regressors for the GLMs. These ratings indicating experience shifts, were compared with neural shifts quantified by a cross-validated Hidden Markov Model. Intersubject Functional connectivity and voxel pattern correlations were computed between the ROIs for quantifying information integration.We show that during naturalistic viewing, our conscious experience emerges as a Fragmented process from the PFC in a tripartite manner (Fig 1b). Moreover, update-driven neural shifts temporally precede experiential shifts and are domain-specific (Fig 1c). This fragmented consciousness gets unified and integrated within the Precuneus (not within the PFC or Hippocampus) in a Multithreaded architecture (Fig 1d). Individuals with different predictions had dissimilar experiences, with divergent neural representations across the sensory and associative cortices, despite the same perceptual information. Fragmentation and Multithreading were replicated in a different cohort, listening to a narrative, suggesting humans utilize a single, abstract predictive mechanism for vision and language inference.These results suggest conscious experience may emerge compositionally from hierarchical, modular architectures requiring both prefrontal generation and parietal integration, challenging the current neural theories of consciousness.