ArtChart: A Benchmark for Faithful Artistic Chart Generation with Integrated Text Rendering

2026-07-17

AI Summary

Existing text-to-image and image editing models struggle to simultaneously ensure geometric accuracy, correct in-chart text rendering, label-token consistency, and stylistic coherence when generating artistic charts, often resulting in distortions, textual hallucinations, or reduced readability. This work proposes ArtChart, a framework that integrates plug-and-play conditional modules to achieve mathematically faithful chart synthesis, enhanced by reinforcement learning driven by OCR accuracy, layout fidelity, and aesthetic rewards, alongside a multi-expert knowledge distillation strategy to jointly optimize textual correctness, structural integrity, and artistic expression. ArtChart represents the first unified approach addressing the triad of mathematical faithfulness, precise text rendering, and artistic stylization, introducing ArtChart-Bench—the first multidimensional benchmark for artistic charts, comprising 2K bilingual prompts—and demonstrating significant superiority over existing open-source baselines across six evaluation dimensions.

Abstract

Artistic charts make data memorable and visually engaging, but generating them faithfully demands simultaneously preserving numerical geometry, rendering exact in-image text, binding labels to correct marks, and maintaining coherent artistic style. Current text-to-image and image editing models frequently fail on these coupled constraints, producing distorted geometries, hallucinated text, misbound labels, or over-stylized marks that undermine readability and mathematical integrity. This paper introduces ArtChart, a framework for artistic chart generation with integrated text rendering, encompassing a task definition, benchmark, and evaluation protocol. This is the first work to simultaneously address mathematically faithful chart synthesis, accurate in-image text rendering, and artistic stylization of chart elements. ArtChart features a chart-specific plug-and-play module conditioned on text-free grayscale chart layouts, ensuring mathematical and logical fidelity. A RL learning strategy with OCR accuracy, layout quality, and aesthetic rewards refines generation, while a multi-expert distillation framework resolves inter-reward conflicts through specialized expert optimization. We construct ArtChart-Bench, a bilingual 2K-prompt benchmark spanning four chart types and diverse label formats. We further design ArtChart-Eval, a six-axis evaluation suite covering mathematical logic, text accuracy, text layout, aesthetics, instruction following, and readability, supporting comparison across T2I, image-editing, controllable-generation, and closed-source API models. Extensive experiments demonstrate that ArtChart consistently outperforms open-source baselines, producing charts that are both visually appealing and math faithful.

Problem

• artistic chart generation
• faithful data visualization
• in-image text rendering
• label-to-mark binding
• mathematical integrity

Innovation

• faithful chart generation
• integrated text rendering
• artistic stylization
• reinforcement learning with OCR reward
• multi-expert distillation