The need to render changes in real time to scenes in a radiosity image is addressed by using parallel processing and distributing the tasks that originate from the scene changes amongst the processors. The mechanism for distributing these tasks is based on hashing the sample points that have been created to random processors. In this way it is ensured that the load balancing mechanism is optimised. In addition the global geometry changes are achieved by global broadcast of data. The combination of these mechanisms enables the system to generate new scenes by utilising most of the sample data from the previous radiosity image.