Empowering LLMs to Understand and Generate Complex Vector Graphics

TL;DR

This paper introduces LLM4SVG, a modular approach that enhances large language models to better understand and generate complex SVG vector graphics by using semantic tokens, structured datasets, and fine-tuning strategies.

Contribution

The paper presents a novel modular architecture with learnable semantic tokens and an automated dataset for instruction following, improving LLMs' capabilities in SVG understanding and generation.

Findings

Developed SVGX-SFT Dataset with 580k SVG instruction examples.

Enhanced LLMs' ability to generate semantically aligned SVGs.

Introduced a modular architecture integrating semantic tags and vector encoders.

Abstract

The unprecedented advancements in Large Language Models (LLMs) have profoundly impacted natural language processing but have yet to fully embrace the realm of scalable vector graphics (SVG) generation. While LLMs encode partial knowledge of SVG data from web pages during training, recent findings suggest that semantically ambiguous and tokenized representations within LLMs may result in hallucinations in vector primitive predictions. Additionally, LLM training typically lacks modeling and understanding of the rendering sequence of vector paths, which can lead to occlusion between output vector primitives. In this paper, we present LLM4SVG, an initial yet substantial step toward bridging this gap by enabling LLMs to better understand and generate vector graphics. LLM4SVG facilitates a deeper understanding of SVG components through learnable semantic tokens, which precisely encode these tokens and their corresponding properties to generate semantically aligned SVG outputs. Using a series of learnable semantic tokens, a structured dataset for instruction following is developed to support comprehension and generation across two primary tasks. Our method introduces a modular architecture to existing large language models, integrating semantic tags, vector instruction encoders, fine-tuned commands, and powerful LLMs to tightly combine geometric, appearance, and language information. To overcome the scarcity of SVG-text instruction data, we developed an automated data generation pipeline that collected our SVGX-SFT Dataset, consisting of high-quality human-designed SVGs and 580k SVG instruction following data specifically crafted for LLM training, which facilitated the adoption of the supervised fine-tuning strategy popular in LLM development.