Add model card, link to code, project page
Browse filesThis PR adds a model card for the paper [LeX-Art: Rethinking Text Generation via Scalable High-Quality Data Synthesis](https://huggingface.co/papers/2503.21749).
README.md
CHANGED
|
@@ -1,17 +1,22 @@
|
|
| 1 |
---
|
| 2 |
-
|
|
|
|
| 3 |
datasets:
|
| 4 |
- X-ART/LeX-10K
|
| 5 |
-
pipeline_tag: text-to-image
|
| 6 |
library_name: diffusers
|
|
|
|
|
|
|
| 7 |
tags:
|
| 8 |
- art
|
| 9 |
- text-rendering
|
| 10 |
-
base_model:
|
| 11 |
-
- Alpha-VLLM/Lumina-Image-2.0
|
| 12 |
---
|
|
|
|
| 13 |
**LeX-Art: Rethinking Text Generation via Scalable High-Quality Data Synthesis**
|
| 14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
We introduce LeX-Art, a comprehensive suite for high-quality text-image synthesis that systematically bridges the gap between prompt expressiveness and text rendering fidelity. Our approach follows a data-centric paradigm, constructing a high-quality data synthesis pipeline based on Deepseek-R1 to curate LeX-10K, a dataset of 10K high-resolution, aesthetically refined 1024$\times$1024 images. Beyond dataset construction, we develop LeX-Enhancer, a robust prompt enrichment model, and train two text-to-image models, LeX-FLUX and LeX-Lumina, achieving state-of-the-art text rendering performance. To systematically evaluate visual text generation, we introduce LeX-Bench, a benchmark that assesses fidelity, aesthetics, and alignment, complemented by Pairwise Normalized Edit Distance (PNED), a novel metric for robust text accuracy evaluation. Experiments demonstrate significant improvements, with LeX-Lumina achieving a 22.16\% PNED gain, and LeX-FLUX outperforming baselines in color (+10.32\%), positional (+5.60\%), and font accuracy (+5.63\%). The codes, models, datasets, and demo are publicly available.
|
| 16 |
|
| 17 |
**Usage of LeX-Lumina:**
|
|
@@ -37,4 +42,8 @@ image = pipe(
|
|
| 37 |
|
| 38 |
).images[0]
|
| 39 |
image.save("lex_lumina_demo.png")
|
| 40 |
-
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
base_model:
|
| 3 |
+
- Alpha-VLLM/Lumina-Image-2.0
|
| 4 |
datasets:
|
| 5 |
- X-ART/LeX-10K
|
|
|
|
| 6 |
library_name: diffusers
|
| 7 |
+
license: mit
|
| 8 |
+
pipeline_tag: text-to-image
|
| 9 |
tags:
|
| 10 |
- art
|
| 11 |
- text-rendering
|
|
|
|
|
|
|
| 12 |
---
|
| 13 |
+
|
| 14 |
**LeX-Art: Rethinking Text Generation via Scalable High-Quality Data Synthesis**
|
| 15 |
|
| 16 |
+
This repository contains the model presented in the paper [LeX-Art: Rethinking Text Generation via Scalable High-Quality Data Synthesis](https://huggingface.co/papers/2503.21749).
|
| 17 |
+
|
| 18 |
+
The abstract of the paper is the following:
|
| 19 |
+
|
| 20 |
We introduce LeX-Art, a comprehensive suite for high-quality text-image synthesis that systematically bridges the gap between prompt expressiveness and text rendering fidelity. Our approach follows a data-centric paradigm, constructing a high-quality data synthesis pipeline based on Deepseek-R1 to curate LeX-10K, a dataset of 10K high-resolution, aesthetically refined 1024$\times$1024 images. Beyond dataset construction, we develop LeX-Enhancer, a robust prompt enrichment model, and train two text-to-image models, LeX-FLUX and LeX-Lumina, achieving state-of-the-art text rendering performance. To systematically evaluate visual text generation, we introduce LeX-Bench, a benchmark that assesses fidelity, aesthetics, and alignment, complemented by Pairwise Normalized Edit Distance (PNED), a novel metric for robust text accuracy evaluation. Experiments demonstrate significant improvements, with LeX-Lumina achieving a 22.16\% PNED gain, and LeX-FLUX outperforming baselines in color (+10.32\%), positional (+5.60\%), and font accuracy (+5.63\%). The codes, models, datasets, and demo are publicly available.
|
| 21 |
|
| 22 |
**Usage of LeX-Lumina:**
|
|
|
|
| 42 |
|
| 43 |
).images[0]
|
| 44 |
image.save("lex_lumina_demo.png")
|
| 45 |
+
```
|
| 46 |
+
|
| 47 |
+
See also:
|
| 48 |
+
* [Project page](https://zhaoshitian.github.io/lexart/)
|
| 49 |
+
* [Code](https://github.com/zhaoshitian/LeX-Art)
|