Parallel planning deals with the problem of finding a sequence of sets of independent actions such that the traditional plan is obtained by ordering actions in these sets. Parallel plans are frequently used by planners based on transformation of the planning problem to a different solving formalism such as SAT (Boolean satisfiability) or CP (Constraint Programming). In this paper we propose a modification of the constraint model for parallel planning based on different encoding of action transitions. Rather than encoding states and their changes via actions, we suggest to encode directly transitions between the actions and omit completely the state variables. This novel encoding improved significantly the runtime of the planner due to faster inference which is demonstrated experimentally using planning domains from International Planning Competition.