Structures of natural plants are complex and difficult to model. This paper proposes a fast visual modeling for plants based on a small set of images, and establishes a reasonable plant model.Based on knowledge about growth patterns of the plant, image segmentation and 3D reconstruction are first performed to construct the plant skeleton (trunk and major branches), from which the remainder of the plant grows. Then the system produces the realistic plant model images based on image synthesis and validation. It is unnecessary to acquire the complex structure (such as the complex production rules of L-systems). The method provides a high degree of control over the final shape by image validation,resulting in realistic reconstruction.