SVGen: Interpretable Vector Graphics Generation with Large Language Models

TL;DR

This paper introduces SVGen, a model that generates accurate and complete SVG graphics from natural language, supported by a large dataset and advanced training techniques, improving efficiency and effectiveness over existing methods.

Contribution

We created SVG-1M, a large-scale SVG dataset with natural language descriptions, and developed SVGen, an end-to-end model that enhances SVG generation with semantic accuracy and structural completeness.

Findings

SVGen outperforms general large models in SVG generation quality.

The dataset SVG-1M enables better training and evaluation.

SVGen achieves higher efficiency in generating SVGs from natural language.

Abstract

Scalable Vector Graphics (SVG) is widely used in front-end development and UI/UX design due to its scalability, editability, and rendering efficiency. However, turning creative ideas into precise vector graphics remains a time-consuming challenge. To address this, we introduce SVG-1M, a large-scale dataset of high-quality SVGs paired with natural language descriptions. Through advanced data augmentation and annotation, we create well-aligned Text to SVG training pairs, including a subset with Chain of Thought annotations for enhanced semantic guidance. Based on this dataset, we propose SVGen, an end-to-end model that generates SVG code from natural language inputs. Our approach ensures semantic accuracy and structural completeness, supported by curriculum learning and reinforcement learning optimization. Experiments show that SVGen outperforms general large models and traditional rendering methods in both effectiveness and efficiency. Code, model, and dataset are available on GitHub.