NeuralSVG is an implicit neural representation method for generating vector graphics from text prompts. Inspired by Neural Radiance Fields (NeRFs), it encodes the entire scene into the weights of a small Multi-Layer Perceptron (MLP) network and optimizes it using Score Distillation Sampling (SDS). This approach employs dropout-based regularization techniques to foster a layered structure in the generated SVG, giving each shape an independent significance within the overall scene. Additionally, its neural representation offers the advantage of inference control, allowing users to dynamically adjust the generated SVG based on the input, such as color and aspect ratio, all from a single learned representation. Extensive qualitative and quantitative evaluations show that NeuralSVG outperforms existing methods in generating structured and flexible SVGs. The model has been jointly developed by researchers from Tel Aviv University and MIT CSAIL, and the code has not been made publicly available yet.