A Survey of Large Language Models for Vector Graphics Generation

Abstract:
[Objective] Systematically sort out the research progress of big language model in the field of vector graphics generation, and clarify its technology evolution path and core research issues. 
[Methods] Through searching the relevant literature in the past five years, the existing research is classified and compared according to the technical paradigm, and a comprehensive analysis is made from the aspects of semantic modeling, geometric representation level and generation framework. 
[Results] The research shows that the field has experienced three stages, from visual language semantic guidance, diffusion model assisted vector generation to structured SVG generation with big language model as the core, and the generation quality, structural consistency and editability have gradually improved.
[Limitations]Existing research still has shortcomings in geometric accuracy control, construction of unified evaluation system and stability in complex design scenarios, and relevant methods rely heavily on high-quality data and computing resources.
[Conclusions] Large language model provides a new research paradigm for vector graphics generation, but its potential in structural modeling and human-computer collaborative design still needs further exploration.