Research map / Image generation / Text-to-image diffusion
TextStyleBrush: Transfer of Text Aesthetics From a Single Example
Praveen Krishnan, Rama Kovvuri, Guan Pang, Boris Vassilev, Tal Hassner
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 |
|---|---|---|---|
| add reconstruction loss l_rec (ablation) | l_D + l_R model · Synth-Paired ablation | FID: 165.41 -> 86.79 | Improved |
| add cyclic reconstruction loss l_cyc (ablation) | l_D + l_R + l_rec model · Synth-Paired ablation | SSIM: 0.4164 -> 0.4434 | Improved |
| add cyclic reconstruction loss l_cyc (ablation) | l_D + l_R + l_rec model · Synth-Paired ablation | FID: 86.79 -> 99.45 | Hurt |
| add typeface perceptual loss l_type (ablation) | l_D + l_R + l_rec + l_cyc model · Synth-Paired ablation | FID: 99.45 -> 79.49 | Improved |
| add typeface perceptual loss l_type (ablation) | l_D + l_R + l_rec + l_cyc model · Synth-Paired ablation | SSIM: 0.4434 -> 0.4174 | Hurt |
| use multi-scale style representation instead of single-scale (ablation) | TSB full losses · Synth-Paired ablation | FID: 87.68 -> 79.49 | Improved |
| add foreground mask learning (ablation) | TSB full losses · Synth-Paired ablation | FID: 90.67 -> 79.49 | Improved |
| add foreground mask learning (ablation) | TSB full losses · Synth-Paired ablation | SSIM: 0.4416 -> 0.4174 | Hurt |
| use TSB instead of SRNet | scene text generation · IC15 word images | recognition accuracy: 23.4 -> 97.6 | Improved |
| use TSB instead of Davis et al. | handwriting generation · IAM words | FID: 104.95 -> 44.68 | Improved |
| use TSB instead of SRNet in a user realism test | scene text generation · User study scene text | percent judged real: 28.64 -> 58.88 | Improved |
| use TSB instead of Davis et al. in a user realism test | handwriting generation · User study handwriting | percent judged real: 37.45 -> 65.76 | 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.
- SWAPText — SwapText: Image Based Texts Transfer in Scenes (CVPR 2020) · 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
- CSGO: Content-Style Composition in Text-to-Image Generation (NeurIPS 2025 · similarity 0.86)
- DEADiff: An Efficient Stylization Diffusion Model with Disentangled Representations (CVPR 2024 · similarity 0.86)
- Zero-Shot Styled Text Image Generation, but Make It Autoregressive (CVPR 2025 · similarity 0.86)
- StyleStudio: Text-Driven Style Transfer with Selective Control of Style Elements (CVPR 2025 · similarity 0.85)
- DreamText: High Fidelity Scene Text Synthesis (CVPR 2025 · similarity 0.85)
- SigStyle: Signature Style Transfer via Personalized Text-to-Image Models (AAAI 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.
