Title: Generating Single-Line Art Through Semantic Optimization

Abstract:

Single-line drawings represent a potent artistic medium, demanding that creators abstract and extract the core essence of their subjects. This paper introduces the inaugural semantics-driven technique for the automatic creation of single-line vector drawings. The generation process is directed by either a descriptive text prompt or a reference image. By employing score distillation sampling to refine the parameters of a Uniform Rational B-Spline (URBS) curve, our method guarantees that the resulting artwork is inherently a single, unbroken stroke. This specific representation offers precise control over detail levels, while supplementary loss functions enable the direction of the final aesthetic style. Our experiments indicate that this approach surpasses current state-of-the-art text-to-image models and optimization pipelines, delivering outcomes that are superior in both visual appeal and adherence to the conventions of continuous line artists. Additionally, the vectorized nature of the generated curves facilitates immediate integration into downstream manufacturing workflows, including wire bending, laser engraving, and embroidery. The source code and project results can be accessed at https://github.com/tanguymagne/SLDgen.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC