Rendering-Aware Reinforcement Learning for Vector Graphics Generation

TL;DR

This paper introduces RLRF, a reinforcement learning approach that improves SVG vector graphic generation by using rendering feedback to enhance the fidelity and efficiency of the generated images.

Contribution

It presents a novel RL method that leverages rendering feedback to train vision-language models for more accurate and coherent SVG generation, addressing limitations of supervised methods.

Findings

RLRF outperforms supervised fine-tuning in SVG quality.

The method produces more faithful and efficient SVGs.

Enhanced structural understanding and generalization in SVG generation.

Abstract

Scalable Vector Graphics (SVG) offer a powerful format for representing visual designs as interpretable code. Recent advances in vision-language models (VLMs) have enabled high-quality SVG generation by framing the problem as a code generation task and leveraging large-scale pretraining. VLMs are particularly suitable for this task as they capture both global semantics and fine-grained visual patterns, while transferring knowledge across vision, natural language, and code domains. However, existing VLM approaches often struggle to produce faithful and efficient SVGs because they never observe the rendered images during training. Although differentiable rendering for autoregressive SVG code generation remains unavailable, rendered outputs can still be compared to original inputs, enabling evaluative feedback suitable for reinforcement learning (RL). We introduce RLRF (Reinforcement Learning from Rendering Feedback), an RL method that enhances SVG generation in autoregressive VLMs by leveraging feedback from rendered SVG outputs. Given an input image, the model generates SVG roll-outs that are rendered and compared to the original image to compute a reward. This visual fidelity feedback guides the model toward producing more accurate, efficient, and semantically coherent SVGs. RLRF significantly outperforms supervised fine-tuning, addressing common failure modes and enabling precise, high-quality SVG generation with strong structural understanding and generalization.