Integrated transportation is one of the most important methods to encourage the modal shift from car to public transportation (PT). However, as most cities have an existing multimodal network, it is difficult to expand the current networks by building more PT routes. Thus, integrating different modes through the optimization of hubs is a cost-efficient way to promote sustainable mobility. This paper develops a bilevel multimodal network design problem based on the collaborative optimization of urban transportation hubs. The upper-level problem is formulated as a mixed-integer nonlinear program to achieve a modal shift from congested subnetworks to underutilized subnetworks to realize a balanced use of the entire network. The decision variables are classified into location-based (hub locations) and route-based (route layouts and frequency setting) ones. The lower-level problem is a generalized modal split/traffic assignment problem (GMS/TAP), which captures the mode choices of all modes in the path set. The GMS/TAP is formulated as a nonlinear optimization problem (NLP) and is solved using a hybrid method of the successive average (MSA) algorithm. A hybrid genetic search with advanced diversity control (HGSADC) is developed to solve the bilevel model, where the exploration of the search space is expanded using the biased fitness function and diversification mechanism. The solution properties of the hybrid MSA and HGSADC are demonstrated in two modified nine-node networks. The model performance is illustrated in a real-size network in Jianye district, Nanjing. 9.2% decrease of travel time, 25.7% increase of service level, and a significant modal shift from car to PT are obtained.