The article presents a new method for creating and editing vector graphics, which are widely used in digital art. The proposed pipeline generates customized vector graphics based on textual prompts, while preserving the properties and layer-wise information of an exemplar SVG. The method uses large pre-trained text-to-image models, and optimizes path parameters using both image-level and vector-level losses. The approach is shown to produce diverse customizations of vector graphics with high quality.
Publication date: 22 Sep 2023
Project Page: https://intchous.github.io/SVGCustomization
Paper: https://arxiv.org/pdf/2309.12302
