In recent years, license plate recognition has become an important research topic in the field of computer vision. However, there are still many challenges that need to be addressed, such as the time-consuming collection and labeling of license plates, the imbalance of data, and the privacy issues associated with license plates. To address these challenges, we propose a new method called Style Preserving Generator (SPG) for license plate image synthesis. This method can replace the content in the source image with arbitrary strings and preserve the style. In addition, we propose a method that uses generated images to train the recognizer, which can achieve better performance than training on real data.