The study of ancient documents is important for humanity to understand history. In Japan, many ancient documents were written in the Kuzushiji font. Unlike modern fonts, Kuzushiji, as a traditional handwritten font, has unique text structures, with strokes often being connected, simplified, or distorted. In recent years, most studies on font generation have focused on modern font generation, and few have focused on the generation of ancient handwritten fonts like Kuzushiji. Furthermore, vector images are resolution-independent, possess high editing flexibility, and have consistent output quality when compared to raster images. To improve the efficiency of digital data reprinting and the preservation of historical documents by taking advantage of vector images, we propose a few-shot training-free font generation method for generating Kuzushiji characters from modern fonts based on vector images. By using only modern fonts and a few Kuzushiji characters as input, this model mitigates dataset imbalance effectively. Our method integrates a differentiable rasterizer, conditional diffusion model, discriminator, and Multi-Stroke Encoder for stroke-level text control, optimized using multiple loss functions. We conducted numerous experiments on the generation of Kuzushiji fonts, and the results demonstrate that the proposed method successfully generates Kuzushiji characters in vector images, achieving better results than previous methods.