As formula recognition models grow in complexity, the demand for handwritten formula data has surged. To address this, formula image generation models have been proposed to convert LaTeX printed formulas into handwritten form. However, the leading formula generation model, FormulaGAN, is limited to generating handwritten formulas without paper texture. This paper introduces a generative model called ‘Paper-Style FormulaGAN’ (PFG). By employing self-attention mechanisms and Vision Transformer, PFG successfully generates handwritten formulas with paper texture, expanding the dataset to enhance the performance of formula recognition models in practical scenarios. Experimental results demonstrate that formulas generated by PFG closely resemble real handwritten formulas compared to FormulaGAN. Moreover, recognition models trained with synthetic data from PFG outperform models trained with data from other methods across multiple evaluation metrics.