Generation is very slow if more than 5 object are defined with depth greater than 2. Maybe we should parallel processing?