Abstract: To achieve semantically controllable SMPL body shape generation, a method is proposed to optimize the SMPL shape parameters using an extended virtual measurement module. First, the feature points are predefined on the SMPL model. Then, the anthropometric measurements are obtained by virtual measurement of the SMPL model in T-pose. Finally, the SMPL shape parameters are optimized by minimizing the objective function. Based on this, the paper further explores the relationship between anthropometric measurements and SMPL shape parameters in the shape space, and establishes a semantically controllable association between the variation of anthropometric measurements and shape parameters to control the changes of single anthropometric measurement of SMPL model by modifying shape parameters, thus solving the problem of fine-grained control of SMPL shape. The experimental results show that the proposed method maintains several anthropometric measurements inputs such as height, perineal height, arm span, chest circumference, waist circumference, and hip circumference as constraints results in the error of the measurements of generated model is smaller than that of the input values, not exceeding 50% of the error of the input value showing by 4 groups of males and 4 groups of females. In addition, this method is also applicable to SMPL-H and SMPL-X models.