StrokeNUWA is an innovative work that explores better visual representations, 'division marks', for vector graphics. These visual marks are semantically rich, naturally compatible with LLMs, and highly compressible. Equipped with division marks, StrokeNUWA significantly outperforms traditional LLM-based and optimization-based methods in various metrics for vector graphics generation tasks. Moreover, StrokeNUWA achieves a 94x speedup in inference time and boasts an outstanding SVG code compression ratio of 6.9% compared to previous methods.