Tel Aviv University and MIT CSAIL have developed NeuralSVG, a groundbreaking technology that generates vector graphics from text prompts. This method allows for dynamic conditioning, enabling the generation of multiple color palettes for a single learned representation. By utilizing an implicit neural representation and Score Distillation Sampling, NeuralSVG encourages a layered structure in the SVG outputs, making them more structured and flexible. Surprisingly, the neural representation also allows for inference-time control, enabling users to adjust aspects like color palette and aspect ratio with a single learned representation. Through extensive evaluations, it has been shown that NeuralSVG outperforms existing methods in generating high-quality vector graphics.
https://sagipolaczek.github.io/NeuralSVG/