A curated collection of methods, datasets, and evaluation tools for Handwritten Text Generation (HTG).
❗ Latest Update: 5 Dec 2025
🚧 Under Construction (updates might be needed)
This repo is a work based on the PhD monograph thesis Handwritten Text Generation with Diffusion Models: Beyond Visual Quality
Konstantina Nikolaidou, 2025 - Advisors: Marcus Liwicki, George Retsinas, Giorgos Sfikas
| Method / Model | Paper (full title) | Venue / Year | Granularity / Setting | Paper / Code Link |
|---|---|---|---|---|
| Alonso et al. | Adversarial Generation of Handwritten Text Images Conditioned on Sequences | ICDAR 2019 | offline, word-level | 📄 Paper 💻 Code (unofficial) |
| ScrabbleGAN | ScrabbleGAN: Semi-Supervised Varying Length Handwritten Text Generation | CVPR 2020 | offline, variable-length words | 📄 Paper 💻 Code |
| GANwriting | GANwriting: Content-Conditioned Generation of Styled Handwritten Word Images | ECCV 2020 | offline, word-level | 📄 Paper 💻 Code |
| Davis et al. | Text and Style Conditioned GAN for the Generation of Offline-Handwriting Lines | BMVC 2020 | offline, line-level | 📄 Paper 💻 Code |
| GANwriting-Line | Content and Style Aware Generation of Text-Line Images for Handwriting Recognition | TPAMI 2021 | offline, line-level | 📄 Paper |
| SmartPatch | SmartPatch: Improving Handwritten Word Imitation with Patch Discriminators | ICDAR 2021 | offline, word-level | 📄 Paper 💻 Code |
| HWT (Handwriting Transformers) | Handwriting Transformers | CVPR 2021 | offline, word/line-level | 📄 Paper 💻 Code |
| HiGAN | HiGAN: Handwriting Imitation Conditioned on Arbitrary-Length Texts and Disentangled Styles | AAAI 2021 | offline, arbitrary length | 📄 Paper 💻 Code |
| JokerGAN | JokerGAN: Memory-Efficient Model for Handwritten Text Generation with Text Line Awareness | ACMM 2021 | offline, word-level | 📄 Paper |
| HiGAN+ | HiGAN+: Handwriting Imitation GAN with Disentangled Representations | ACM TOG 2022 | offline, word-level | 📄 Paper 💻 Code |
| SLOGAN | SLOGAN: Handwriting Style Synthesis for Arbitrary-Length and Out-of-Vocabulary Text | IEEE Transactions on Neural Networks and Learning Systems 2022 |
offline, arbitrary-level | 📄 Paper |
| VATr | Handwritten Text Generation From Visual Archetypes | CVPR 2023 | offline, word-level | 📄 Paper 💻 Code |
| TextStyleBrush | TextStyleBrush: Transfer of Text Aesthetics from a Single Example | TPAMI 2023 | offline, in-the-wild, word-level | 📄 Paper |
| JokerGAN++ | Handwritten Text Generation with Character-specific Encoding for Style Imitation | ICDAR 2023 | offline, line-level | 📄 Paper |
| VATr++ | VATr++: Choose Your Words Wisely for Handwritten Text Generation | TPAMI 2024 | offline, word/line-level | 📄 Paper 💻 Code |
| — | — | — | — | — |
| Method / Model | Paper (full title) | Venue / Year | Granularity / Setting | Paper / Code Link |
|---|---|---|---|---|
| CTIG-DM | Conditional Text Image Generation with Diffusion Models | CVPR 2023 | offline, scene-text, word-level, non-latent | 📄 Paper |
| WordStylist | WordStylist: Styled Verbatim Handwritten Text Generation with Latent Diffusion Models | ICDAR 2023 | offline, word-level, latent | 📄 Paper 💻 Code |
| DiffusionPen | DiffusionPen: Few-Shot Handwritten Text Generation with Hybrid Style Encoding | ECCV 2024 | offline, in-the-wild, one-shot, few-shot, word (+line), latent | 📄 Paper 💻 Code |
| One-DM | One-DM: One-Shot Diffusion Mimicker for Handwritten Text Generation | ECCV 2024 | offline, one-shot, non-latent, rendered content condition | 📄 Paper 💻 Code |
| DiffWord | Word-Diffusion: Diffusion-Based Handwritten Text Word Image Generation | ICPR 2024 | offline, word-level | 📄 Paper |
| Brandenbusch | Semi-Supervised Adaptation of Diffusion Models for Handwritten Text Generation | arXiv 2024 | offline, word-level, few-shot, CFG | 📄 Paper |
| Zero-Shot Paragraph DM | Zero-Shot Paragraph-level Handwriting Imitation with Latent Diffusion Models | IJCV 2025 | offline, paragraph-level | 📄 Paper 💻 Code |
| DOG | Dual Orthogonal Guidance for Robust Diffusion-based Handwritten Text Generation | ICCV Workshops 2025 | diffusion guidance, offline, in-the-wild | 📄 Paper |
| Method / Model | Paper (full title) | Venue / Year | Granularity / Setting | Paper / Code Link |
|---|---|---|---|---|
| Emuru | Zero-Shot Styled Text Image Generation, but Make It Autoregressive | CVPR 2025 | font-based, synthetic, line-level, autoregressive | 📄 Paper 💻 Code |
| Eruku | Zero-Shot Styled Text Image Generation, but Make It Autoregressive | WACV 2026 | font-based, synthetic, line-level, autoregressive | 📄 Paper 💻 Code |
| Method / Model | Paper (full title) | Venue / Year | Granularity / Setting | Paper / Code Link |
|---|---|---|---|---|
| Cross-VAE | Modality Conversion of Handwritten Patterns by Cross Variational Autoencoders | ICDAR 2019 | online/offline conversion, character-level | 📄 Paper |
| DeepWriteSYN | DeepWriteSYN: On-line Handwriting Synthesis via Deep Short-Term Representations | AAAI 2021 | online, character/short-sequence | 📄 Paper |
| Emuru-VAE | Zero-Shot Styled Text Image Generation, but Make It Autoregressive | CVPR 2025 | font-based, writer/HTR VAE pre-training on synth lines | 📄 Paper 💻 Code |
| Dataset | Paper | Year | Description | Links |
|---|---|---|---|---|
| IAM Handwriting Database | The IAM-database: an English sentence database for offline handwriting recognition – Marti & Bunke | 2002 | 1.5K pages, 82K text lines, 115K words of unconstrained English handwriting from 657 writers. Standard benchmark for Latin-script HTR and HTG. | 📄 Paper 🗂️ Dataset |
| CVL Handwriting Database | CVL-DataBase: An Off-Line Database for Writer Retrieval, Writer Identification and Word Spotting – Kleber et al. | 2013 | English/German handwritten texts from 310 writers, supporting writer identification, retrieval and word-level analysis. | 📄 Paper 🗂️ Dataset |
| RIMES | RIMES evaluation campaign for handwritten mail processing – Augustin et al. | 2006 | Large-scale French handwriting database collected from postal mail, with page- and line-level annotations for HTR. | 📄 Paper 🗂️ Dataset |
| GNHK | GNHK: a dataset for English handwriting in the wild – Lee et al. | 2021 | 687 camera-captured pages with 39K text instances, 9K lines and 173K characters. Focused on in-the-wild handwritten text under real-world degradations. | 📄 Paper 🗂️ Dataset |
| Imgur5K | TextStyleBrush: Transfer of Text Aesthetics From a Single Example – Krishnan et al. | 2023 | ~5K images and ~135K word instances of diverse “in-the-wild” handwriting sourced from Imgur, used to evaluate one-shot style transfer and robust HTG. | 📄 Paper 🗂️ Dataset |
| Dataset | Paper | Year | Description | Links |
|---|---|---|---|---|
| IIIT-HWS | Generating Synthetic Data for Text Recognition – Krishnan & Jawahar | 2016 | ~9M synthetic handwritten word images rendered with handwriting-style fonts and heavy augmentation, widely used for pretraining HTR/word-spotting models. | 📜 Paper 📂 Dataset |
| Font² | Evaluating synthetic pre-training for handwriting processing tasks – Pippi et al. | 2023 | Large-scale synthetic word corpus rendered with thousands of calligraphic/handwriting fonts, designed for supervised pretraining of style encoders and handwriting analysis. | 📜 Paper 📂 Dataset |
If you find this work useful for your research, please cite:
@phdthesis{nikolaidou2025handwritten,
title={Handwritten Text Generation with Diffusion Models: Beyond Visual Quality},
author={Nikolaidou, Konstantina},
year={2025},
school={Lule{\aa} University of Technology}
}