SVGen: Interpretable Vector Graphics Generation with Large Language Models

2025-08-06

AI Summary

This work addresses three key challenges in natural language-to-SVG generation: poor semantic alignment, low structural integrity, and weak interpretability. To tackle these, we propose SVGen—a novel end-to-end generative model—and introduce SVG-1M, a large-scale, high-quality dataset specifically curated for this task. Methodologically, we innovatively incorporate chain-of-thought (CoT) annotations and curriculum learning, integrated with large language model fine-tuning, synthetic data augmentation, and reinforcement learning optimization to enhance joint text understanding and SVG syntax generation. Experimental results demonstrate that SVGen significantly outperforms both general-purpose large language models and conventional rendering-based approaches across accuracy, structural validity, and rendering fidelity. We publicly release the model, source code, and dataset to foster reproducible research. This work establishes a new paradigm for UI design automation and interpretable, AI-driven vector graphic generation.

Technology Category

• Natural Language Processing: Generation
• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search

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.

Problem

• Turning creative ideas into precise vector graphics efficiently
• Generating SVG code from natural language inputs accurately
• Ensuring semantic accuracy and structural completeness in SVG generation

Innovation

• Large-scale SVG dataset with text descriptions
• End-to-end SVG generation from natural language
• Curriculum and reinforcement learning optimization