NIV: Neural Axis Variations for Variable Font Generation

This dataset contains over one million variation tuples derived from variable Google Fonts, used for training the NIV (Neural Axis Variations) model for automatic variable font generation.

• Paper: NIV: Neural Axis Variations for Variable Font Generation
• Project Page: https://ndvbd.github.io/NIV/
• Repository: https://github.com/ndvbd/NIV

Dataset Description

The dataset comprises per-point displacements for glyph outlines across various design axes such as weight (wght), width (wdth), slant (slnt), and optical size (opsz). It is designed to enable neural models to predict continuous geometric variations for vector glyphs, allowing for the conversion of static fonts into functional variable fonts.

Sample Usage

As per the official repository, you can download the prepared dataset using the huggingface_hub CLI:

pip install -U huggingface_hub
hf download ndvb/NIV \
  --repo-type dataset \
  --local-dir dataset

Citation

If you find this work or dataset useful in your research, please consider citing:

@article{benedek2026niv,
  title={NIV: Neural Axis Variations for Variable Font Generation},
  author={Benedek, Nadav and Shamir, Ariel and Fried, Ohad},
  journal={arXiv preprint arXiv:2606.05261},
  year={2026}
}