Research map / Image generation / Text-to-image diffusion
FonTS: Text Rendering With Typography and Style Controls
Wenda Shi, Yiren Song, Dengming Zhang, Jiaming Liu, Xingxing Zou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| fine-tune only joint text attention (5% params) rather than text+image attention (ablation) | TC-FT on Flux.1-dev · BTR-bench | Word-Acc: 31.00 -> 55.00 | Improved |
| fine-tune only joint text attention rather than text+image attention (ablation) | TC-FT on Flux.1-dev · BTR-bench | OCR-Acc: 77.92 -> 82.85 | Improved |
| fine-tune T5 text encoder with new tokens (ablation) | TC-FT on Flux.1-dev · BTR-bench | OCR-Acc: 82.85 -> 0 | Hurt |
| train new tokens only without fine-tuning attention (ablation) | TC-FT on Flux.1-dev · BTR-bench | OCR-Acc: 82.85 -> 66.52 | Hurt |
| use enclosing token pair instead of single modifier token (ablation) | ETC-tokens · BTR-bench | Word-Acc: 32.00 -> 55.00 | Improved |
| use ETC-tokens instead of natural-language bold instruction (ablation) | ETC-tokens · BTR-bench | Word-Acc: 25.00 -> 55.00 | Improved |
| add second-phase Art-FT on SC-artext to style adapter (ablation) | SCA · ATR-bench | CLIP-I: 60.07 -> 65.12 | Improved |
| add TC-FT backbone fine-tuning (ablation) | SCA without Art-FT · ATR-bench | OCR-Acc: 28.89 -> 65.39 | Improved |
| add TC-FT backbone fine-tuning (ablation) | SCA without Art-FT · ATR-bench | CLIP-I: 60.07 -> 58.09 | Hurt |
| apply SCA to both MM-DiT and Single-DiT instead of MM-DiT only (ablation) | SCA · ATR-bench | Avg: 57.34 -> 62.17 | Improved |
| FonTS vs Glyph-ByT5 on OCR accuracy in basic text rendering | BTR · BTR-bench | OCR-Acc: 96.36 -> 82.85 | Hurt |
| use FonTS instead of Flux for scene text | Flux.1-dev · MARIO-bench | OCR-Acc: 24.17 -> 53.57 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TextDiffuser-2 — TextDiffuser-2: Unleashing the Power of Language Models for Text Rendering (ECCV 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Type-R: Automatically Retouching Typos for Text-to-Image Generation (CVPR 2025 · similarity 0.88)
- FontDiffuser: One-Shot Font Generation via Denoising Diffusion with Multi-Scale Content Aggregation and Style Contrastive Learning (AAAI 2024 · similarity 0.86)
- How Control Information Influences Multilingual Text Image Generation and Editing? (NeurIPS 2024 · similarity 0.86)
- TextCtrl: Diffusion-based Scene Text Editing with Prior Guidance Control (NeurIPS 2024 · similarity 0.85)
- Ctrl-X: Controlling Structure and Appearance for Text-To-Image Generation Without Guidance (NeurIPS 2024 · similarity 0.85)
- StyleStudio: Text-Driven Style Transfer with Selective Control of Style Elements (CVPR 2025 · similarity 0.85)
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.
