Yesianrohn/artistic-fonts

Artistic Fonts Collection

Artistic Fonts is a large-scale collection of artistic / decorative display fonts used by the SynthWordArt rendering engine to build the WATER-T subset of WATER-S in the paper "Advancing WordArt-Oriented Scene Text Recognition: Datasets and Methods" (ECCV 2026).

These open-source artistic fonts were collected from the internet and cover a wide spectrum of artistic styles. They replace the commonly used standard fonts in classical text-rendering pipelines (SynthText / SynthTIGER) and are a key ingredient for synthesizing diverse, stylized WordArt training data.

• 📄 Paper (arXiv): https://arxiv.org/abs/2606.24484
• 💻 Code: https://github.com/YesianRohn/WATER
• 📦 Main dataset (WATER-Data): https://huggingface.co/datasets/Yesianrohn/WATER-Data
• 🏋️ Model weights: https://huggingface.co/Yesianrohn/WATERec-Models

---

Statistics

Item	Count
Total fonts	~11,213
TrueType (.ttf)	6,432
OpenType (.otf)	4,781

---

Data Structure

Stored as Hugging Face native Parquet shards. The font binary is embedded directly so the dataset is self-contained.

artistic-fonts/
├── data/
│   ├── train-00000-of-00004.parquet
│   ├── train-00001-of-00004.parquet
│   ├── train-00002-of-00004.parquet
│   └── train-00003-of-00004.parquet   # ~1.03 GB total
└── README.md

Each record has the following fields:

Column	Type	Description
font_name	string	Font family name
file_name	string	Original file name (with extension)
format	string	ttf or otf
font_bytes	binary	Raw bytes of the font file
num_chars	int64	Number of supported characters
supported_chars	string	Supported-character coverage string

---

Usage

from datasets import load_dataset

ds = load_dataset("Yesianrohn/artistic-fonts", split="train")
print(ds[0])
# {'font_name': 'a Ablasco', 'file_name': 'a Ablasco.ttf', 'format': 'ttf',
#  'font_bytes': b'\x00\x01\x00\x00...', 'num_chars': 95, 'supported_chars': '...'}

# Write a font back to disk
row = ds[0]
with open(row["file_name"], "wb") as f:
    f.write(row["font_bytes"])

# Filter by format
ttf_fonts = ds.filter(lambda x: x["format"] == "ttf")
otf_fonts = ds.filter(lambda x: x["format"] == "otf")

---

Intended Use

For research on artistic-text data synthesis (rendering WordArt with diverse fonts) and typography-related tasks.

---

License

The fonts are collected from various free-font websites. Before any commercial use, you must verify the individual license of each font. This repository is distributed under an aggregate "other" (per-font) license and does not override the original font licenses.

---

Citation

If you use this collection in your research, please cite our paper:

@inproceedings{water2026eccv,
  title     = {Advancing WordArt-Oriented Scene Text Recognition: Datasets and Methods},
  author    = {Ye, Xingsong and Du, Yongkun and Zhang, Jiaxin and Zhang, Haojie and Sun, Chong and Li, Chen and Lyu, Jing and Chen, Zhineng},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2026}
}