DiffInk: Glyph- and Style-Aware Latent Diffusion Transformer for Text to Online Handwriting Generation

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Pan, Wei, He, Huiguo, Cheng, Hiuyi, Shi, Yilin, Jin, Lianwen
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918362460717056
author Pan, Wei
He, Huiguo
Cheng, Hiuyi
Shi, Yilin
Jin, Lianwen
author_facet Pan, Wei
He, Huiguo
Cheng, Hiuyi
Shi, Yilin
Jin, Lianwen
contents Deep generative models have advanced text-to-online handwriting generation (TOHG), which aims to synthesize realistic pen trajectories conditioned on textual input and style references. However, most existing methods still primarily focus on character- or word-level generation, resulting in inefficiency and a lack of holistic structural modeling when applied to full text lines. To address these issues, we propose DiffInk, the first latent diffusion Transformer framework for full-line handwriting generation. We first introduce InkVAE, a novel sequential variational autoencoder enhanced with two complementary latent-space regularization losses: (1) an OCR-based loss enforcing glyph-level accuracy, and (2) a style-classification loss preserving writing style. This dual regularization yields a semantically structured latent space where character content and writer styles are effectively disentangled. We then introduce InkDiT, a novel latent diffusion Transformer that integrates target text and reference styles to generate coherent pen trajectories. Experimental results demonstrate that DiffInk outperforms existing state-of-the-art (SOTA) methods in both glyph accuracy and style fidelity, while significantly improving generation efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2509_23624
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DiffInk: Glyph- and Style-Aware Latent Diffusion Transformer for Text to Online Handwriting Generation
Pan, Wei
He, Huiguo
Cheng, Hiuyi
Shi, Yilin
Jin, Lianwen
Computer Vision and Pattern Recognition
Deep generative models have advanced text-to-online handwriting generation (TOHG), which aims to synthesize realistic pen trajectories conditioned on textual input and style references. However, most existing methods still primarily focus on character- or word-level generation, resulting in inefficiency and a lack of holistic structural modeling when applied to full text lines. To address these issues, we propose DiffInk, the first latent diffusion Transformer framework for full-line handwriting generation. We first introduce InkVAE, a novel sequential variational autoencoder enhanced with two complementary latent-space regularization losses: (1) an OCR-based loss enforcing glyph-level accuracy, and (2) a style-classification loss preserving writing style. This dual regularization yields a semantically structured latent space where character content and writer styles are effectively disentangled. We then introduce InkDiT, a novel latent diffusion Transformer that integrates target text and reference styles to generate coherent pen trajectories. Experimental results demonstrate that DiffInk outperforms existing state-of-the-art (SOTA) methods in both glyph accuracy and style fidelity, while significantly improving generation efficiency.
title DiffInk: Glyph- and Style-Aware Latent Diffusion Transformer for Text to Online Handwriting Generation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.23624