Fudan University & StepFun · Text-to-Image

OmniSVG

A unified framework for end-to-end multimodal SVG generation that tokenizes SVG commands into discrete tokens and leverages pre-trained Vision-Language Models to generate complex vector graphics across multiple modalities (Text-to-SVG, Image-to-SVG, Character-Reference SVG). Introduces MMSVG-2M, a 2M-asset multimodal dataset with standardized evaluation protocols for SVG synthesis tasks.