Graphical User Interface (GUI) design is currently shifting from designing GUIs composed of standard widgets to designing GUIs relying on more natural interactions and ad hoc widgets. This shift is meant to support the advent of GUIs providing users with more adapted and natural interactions, and the support of new input devices such as multi-touch screens. Standard widgets (e.g. buttons) are more and more replaced by ad hoc ones (e.g. the drawing area of graphical editors), and interactions are shifting from mono-event (e.g. button pressures) to multi-event interactions (e.g. multi-touch and gesture-based interactions). As a consequence, the current GUI model-based testing approaches, which target event-based systems, show their limits when applied to test such new advanced GUIs. The work introduced in this paper establishes three contributions: a precise analysis of the reasons of these current limits; a proposition to tackle the identified limits by leveraging the Malai GUI specification language and by proposing the concept of interaction-action-flow graph; feedback from two use cases, an industrial project and an open-source application, where the proposed approach has been applied.