This application provides a generative flow model training method, an action prediction method, and an apparatus, to implement training of a generative flow model with a directed acyclic graph or a cyclic graph, so as to be more widely applied to a plurality of artificial intelligence scenarios. The method includes: obtaining a plurality of state nodes in a generative flow model; then obtaining a value of a loss function based on the plurality of state nodes, where the value of the loss function includes a value obtained by fusing a value of a flow matching function and a value of a constraint function, each state node has an input flow and an output flow, the value of the flow matching function includes a value calculated based on the input flow and the output flow of each state node in the plurality of state nodes.