Render farm is widely used in movie industry to solve the long rendering time problem. By parallel computing rendering jobs, render farm can speedup the rendering process in a scalable way. In this paper, we present an efficient design of render farm system name DRFarm in distributed environment. The most important feature of the system is the capacity aware task scheduling strategy. We first introduce the hierarchy tasks subdivision method which ensures flexible merging and dividing tasks. By carefully grouping tasks and dynamically assign them in different modes, the overall parallel rendering time can be reduced by exploiting coherence comparing to conventional methods. Furthermore, to adopt various rendering jobs from different locations, we design a general rendering service interface with unified job definition.