Optimal fractional differential order can be only specified manually by experiments. In order to significantly save time of manually seeking optimal fractional differential order, a novel approach is studied and proposed, which is based on window size, G-L formula, image gradient and visual property to automatically generate fractional differential order. The fractional differential operator masks are designed and realized by employing this order. The evaluation parameters of image texture, such as information entropy and average gradient were used for quantitative analysis and experimental verification. The results show that, for any grey image, this method can obtain a successive enhancement result, which approximates the results of optimal fractional differential order and more satisfies human visual sense. It is an effective approach to enhance image texture.