From Seeing to Predicting: A Vision-Language Framework for Trajectory Forecasting and Controlled Video Generation | Synapse