Beyond Patches: Global-aware Autoregressive Model for Multimodal Few-Shot Font Generation

2026-01-04

arXiv.org

AI Summary

This work proposes GAR-Font, a novel framework addressing the challenge of simultaneously preserving glyph structural integrity and stylistic consistency in few-shot font generation. GAR-Font introduces, for the first time, a global-aware mechanism into autoregressive font synthesis, employing a global-aware tokenizer to jointly model local details and global style. It further incorporates a lightweight language-style adapter that enables text-driven style control without requiring large-scale multimodal pretraining. A post-optimization refinement pipeline is integrated to significantly enhance output quality. Experimental results demonstrate that GAR-Font outperforms existing methods in both structural accuracy and style consistency, exhibiting particularly strong capability in faithfully reproducing target styles under textual guidance.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Natural Language Processing: Generation
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Manual font design is an intricate process that transforms a stylistic visual concept into a coherent glyph set. This challenge persists in automated Few-shot Font Generation (FFG), where models often struggle to preserve both the structural integrity and stylistic fidelity from limited references. While autoregressive (AR) models have demonstrated impressive generative capabilities, their application to FFG is constrained by conventional patch-level tokenization, which neglects global dependencies crucial for coherent font synthesis. Moreover, existing FFG methods remain within the image-to-image paradigm, relying solely on visual references and overlooking the role of language in conveying stylistic intent during font design. To address these limitations, we propose GAR-Font, a novel AR framework for multimodal few-shot font generation. GAR-Font introduces a global-aware tokenizer that effectively captures both local structures and global stylistic patterns, a multimodal style encoder offering flexible style control through a lightweight language-style adapter without requiring intensive multimodal pretraining, and a post-refinement pipeline that further enhances structural fidelity and style coherence. Extensive experiments show that GAR-Font outperforms existing FFG methods, excelling in maintaining global style faithfulness and achieving higher-quality results with textual stylistic guidance.

Problem

• Few-Shot Font Generation
• Autoregressive Model
• Global-aware Tokenization
• Multimodal Style Representation
• Stylistic Fidelity

Innovation

• global-aware tokenization
• autoregressive font generation
• multimodal few-shot learning
• language-guided style control
• post-refinement pipeline