BoundInk: Boundary-Aware Online Handwriting Generation
Jinsu Shin, Sungeun Hong, JinYeong Bak
Sungkyunkwan University
Published September 20, 2026 · on the feed September 28, 2026 · ▲ 6 · arXiv 2604.02103
The abstract
Realistic online handwriting depends not only on individual character shapes, but also on how a writer connects, spaces, and aligns adjacent characters. Existing methods rely primarily on long-range sequence modeling and capture these inter-character behaviors only implicitly. This often produces plausible glyphs accompanied by broken cursive joins, inconsistent spacing, or writer-inconsistent transitions. We introduce BoundInk, a writer-conditioned framework that treats inter-character boundaries as explicit generation units. By jointly modeling local transitions and surrounding text context, BoundInk preserves writer-specific glyph appearance while improving connectivity and spacing across complete text lines. We further introduce a boundary-aware evaluation framework that directly assesses cursive continuity and spatial relationships between characters beyond conventional trajectory similarity. Across three benchmark-matched settings, BoundInk improves all applicable boundary-quality measures and reduces normalized dynamic time warping by 17.6--47.8%. In blind human evaluations, BoundInk outputs are preferred in 78.0--82.6% of valid criterion-wise judgments.
The Brief
2:04A short podcast summary of the paper. Written and voiced by AI.
Transcript
BoundInk. Boundary-Aware Online Handwriting Generation. By Jinsu Shin and two others, from Sungkyunkwan University. Published on the twentieth of September. Its code repository is worked out: about forty-five source files, with a readme.
Handwriting looks real for reasons that go beyond the shape of each letter. It also depends on how a writer connects one character to the next, how they space them, and how they align neighbours along the line. Existing online handwriting generators rely mainly on long-range sequence modeling, and they capture those inter-character behaviours only implicitly. The result is a familiar kind of near miss: glyphs that look plausible on their own, joined by broken cursive connections, uneven spacing, or transitions that do not match the writer.
BoundInk makes the space between characters a first-class object. It is a writer-conditioned framework that treats inter-character boundaries as explicit generation units, rather than leaving them to emerge from a long sequence model.
Each boundary is modeled jointly from two sources: the local transition itself, and the surrounding text context. Doing both at once lets BoundInk keep the writer-specific appearance of individual glyphs while improving connectivity and spacing across a complete line of text. Because conventional trajectory similarity does not directly measure these properties, the authors also introduce a boundary-aware evaluation framework. It assesses cursive continuity and the spatial relationships between characters as things in their own right.
The comparison spans three benchmark-matched settings. In all three, BoundInk improves every applicable boundary-quality measure, and it reduces normalized dynamic time warping by between seventeen point six percent and forty-seven point eight percent. People were asked too. In blind human evaluations, BoundInk's outputs are preferred in between seventy-eight percent and eighty-two point six percent of valid criterion-wise judgments.
One way to hold this paper in mind is as a change of unit. The letter used to be the thing generated and the join was left to chance. Here the join is generated deliberately, and the evaluation is rebuilt so that joins are what gets measured.
Paper Quest · 7 chapters · about 35 min
Play this paper
Learn it by doing: make the calls the authors made, guess the result before you see it, and defend it in a viva graded on your own words. Chapter 1 is right here.
Chapter 1 🧩 · The problem
Every letter looks right, yet the line looks wrong. That is where this quest begins.
You are building a model that writes handwriting as pen trajectories in a given person's style. Each letter on its own looks right.
The road ahead
- 1 · The problem ~6 min A handwriting model can draw every letter well and still produce a line that looks wrong, because the joins and gaps between letters fail.
- 2 · What came before ~3 min
- 3 · The key idea ~9 min BoundInk treats the boundary between neighbouring characters as something to generate on purpose, with gated sentence context and direct checks of joins and spacing.
- 4 · Choose your own research ~3 min
- 5 · How it works ~3 min
- 6 · The experiment ~6 min You’ll guess: In blind human comparisons against earlier handwriting generators, in roughly what share of valid judgments were BoundInk's outputs preferred?
- 7 · The viva ~5 min Three questions from the examiners, answered in your own words. Win it for the paper’s trophy.
Free, with Google or an email address. Your stars and progress are kept.