NeuralSVG: Implicit Neural Representation for Text-to-Vector Generation
2025-01-08
NeuralSVG generates vector graphics from text prompts, utilizing an implicit neural representation (similar to NeRFs) encoded into a small MLP network and optimized using Score Distillation Sampling (SDS). A dropout-based regularization technique ensures ordered, editable shapes with a layered structure. The model supports dynamic control over aspects like background color and aspect ratio, all from a single learned representation. Experiments show NeuralSVG outperforms existing methods in generating structured and flexible SVGs.
Read more