Beyond Flat Text: Dual Self-inherited Guidance for Visual Text Generation
Minxing Luo 1, Zixun Xia Beyond Flat Text: Dual Self-inherited Guidance for Visual Text Generation 1, Liaojun Chen 1, Zhenhang Li 2, Weichao Zeng 2, Jianye Wang 1 · 2025
Read Beyond Flat Text: Dual Self-inherited Guidance for Visual Text Generation by Minxing Luo 1, Zixun Xia Beyond Flat Text: Dual Self-inherited Guidance for Visual Text Generation 1, Liaojun Chen 1, Zhenhang Li 2, Weichao Zeng 2, Jianye Wang 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-05892.txt.
Description
Luo and nine co-authors generated slanted and curved text in images for this paper from twenty twenty-five. Words on cans, banners, and badges rarely sit flat. Diffusion models botch such layouts, warping letters and clashing with backgrounds, because training data skew flat. Their framework, STGen, needs no retraining of the base model.
One branch borrows the clean meaning of flat text the model already makes well. That guidance repairs the semantics of hard layouts. A second branch injects the letter shapes from a glyph image to hold structure firm. A merging step blends the guides for a harmonious result. Tests across many layouts showed top accuracy and quality.
Catalog record
| CREATOR | Minxing Luo 1, Zixun Xia Beyond Flat Text: Dual Self-inherited Guidance for Visual Text Generation 1, Liaojun Chen 1, Zhenhang Li 2, Weichao Zeng 2, Jianye Wang 1 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-05892