In our work, we developed a pattern-producing neural network that we named Transitional Pattern Producing Networks (TPPN). The TPPN generates soft robot morphologies much faster than existing methods. Our TPPN architecture uses only sine activation function with trainable frequency and phase. It uses a Boolean mask that turns neural connections on/off during evolution. In addition to running much faster than existing methods, the TPPN generates morphologies resembling biological patterns such as bilateral symmetry and smoothly transitions between patterns during successive evolutionary runs. The ability to smoothly transition between patterns during successive generations has been a major challenge for existing pattern-producing models, especially in robotics.