Research map / Multi-agent systems and games / LLM agents
OneFont: A Unified Agent for End-to-End Font Creation
Yingxin Lai, Yufei Liu, Guoqing Yang, Jiaxing Chai, Zhiming Luo, Shaozi Li
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes OneFont, a dialogue-driven unified agent that handles both glyph synthesis and refinement for end-to-end font creation, using a Font with Thought paradigm.
Problem
Font creation is a laborious trial-and-error workflow; practitioners must iterate manually across generation and refinement steps.
Approach
Interprets user intent through free-form dialogue and integrates glyph synthesis and refinement modules in one framework; the Font with Thought (FwT) paradigm reframes font design as a reasoning task where the model plans the design before acting.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Graph-Based Planner (ablation) | OneFont · FontBench test | Font ACC (%): 51.5 -> 53.8 | Improved |
| add Graph-Based Planner (ablation) | OneFont · FontBench test | FID: 29.4 -> 27.1 | Improved |
| add StrokeMicro Adjuster on top of GBP (ablation) | OneFont · FontBench test | Font ACC (%): 53.8 -> 55.7 | Improved |
| add Preference Alignment (GRPO) instead of SA (ablation) | OneFont · FontBench test | Font ACC (%): 55.7 -> 56.4 | Improved |
| combine GBP, SA and PA (ablation) | OneFont · FontBench test | FID: 29.4 -> 23.2 | Improved |
| add preference alignment after SFT (ablation) | OneFont · Font generation (P, All) | LPIPS: 0.678 -> 0.645 | Improved |
| add preference alignment after SFT (ablation) | OneFont · Font generation (D, All) | SSIM: 0.294 -> 0.326 | Improved |
| fine-tune VLM planner on FontBench vs closed-source LLMs | OneFont · FontBench test | Model ACC (%): 60.1 -> 81.7 | Improved |
| OneFont vs MetaDesigner in user study | OneFont · User study | Aesthetics: 61.5 -> 73.1 | Improved |
Closest papers by meaning
- Font-Agent: Enhancing Font Understanding with Large Language Models (CVPR 2025 · similarity 0.87)
- IF-Font: Ideographic Description Sequence-Following Font Generation (NeurIPS 2024 · similarity 0.85)
- VQ-FONT: Few-Shot Font Generation with Structure-Aware Enhancement and Quantization (AAAI 2024 · similarity 0.85)
- Generate Like Experts: Multi-Stage Font Generation by Incorporating Font Transfer Process into Diffusion Models (CVPR 2024 · similarity 0.84)
- FontDiffuser: One-Shot Font Generation via Denoising Diffusion with Multi-Scale Content Aggregation and Style Contrastive Learning (AAAI 2024 · similarity 0.84)
- MetaDesigner: Advancing Artistic Typography through AI-Driven, User-Centric, and Multilingual WordArt Synthesis (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
