Improve model card for TextPecker Qwen-Image adapter

Hi! I'm Niels from the Hugging Face community science team. This PR improves the model card for the TextPecker-optimized Qwen-Image model.

Changes include:
- Added the `text-to-image` pipeline tag.
- Added the `apache-2.0` license based on the official repository.
- Linked the research paper and GitHub repository.
- Structured the description and usage information for better readability.
- Retained the existing sample usage code which demonstrates how to load the LoRA weights using `peft` and `diffusers`.

README.md

@@ -1,21 +1,30 @@
 ---
 base_model: Qwen/Qwen-Image
 library_name: peft
+license: apache-2.0
+pipeline_tag: text-to-image
+tags:
+- lora
+- visual-text-rendering
+- text-to-image
 ---
-# Model Card for Model ID
 
-<!-- Provide a quick summary of what the model is/does. -->
-This model is trained using Flow-GRPO with LoRA. We provide only the LoRA weights here, so you will need to download the Qwen-Image base model first.
+# QwenImage-TextPecker-SQPA
 
-## Model Details
+This model is a LoRA adapter for [Qwen/Qwen-Image](https://huggingface.co/Qwen/Qwen-Image), trained using the TextPecker structural anomaly perceptive RL strategy. TextPecker is designed to enhance Visual Text Rendering (VTR) by quantifying and rewarding structural anomalies like distortion and misalignment.
 
-### Model Sources
+- **Paper:** [TextPecker: Rewarding Structural Anomaly Quantification for Enhancing Visual Text Rendering](https://arxiv.org/abs/2602.20903)
+- **Repository:** [https://github.com/CIawevy/TextPecker](https://github.com/CIawevy/TextPecker)
+- **Base Model:** [Qwen/Qwen-Image](https://huggingface.co/Qwen/Qwen-Image)
 
-<!-- Provide the basic links for the model. -->
+## Model Description
+
+Visual Text Rendering (VTR) remains a critical challenge in text-to-image generation. Even advanced models frequently produce text with structural anomalies. TextPecker addresses this using a structural anomaly perceptive RL strategy that works with any text-to-image generator. When applied to Qwen-Image, it yields significant gains in structural fidelity and semantic alignment for text rendering.
+
+## Usage
+
+This repository provides only the LoRA weights (SQPA). You must download the `Qwen-Image` base model first to use this adapter.
 
-- **Repository:** https://github.com/CIawevy/TextPecker/tree/main
-- **Paper:** https://www.arxiv.org/pdf/2602.20903
-## Uses
 ```python
 import os
 import torch
@@ -27,8 +36,8 @@ os.environ["PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION"] = "python"
 os.environ["DIFFUSERS_DISABLE_NATIVE_ATTENTION"] = "1"
 
 def load_model(model_path, ckpt_path=None, use_lora=True):
-    torch_dtype = torch.bfloat16
-    device = "cuda"
+    torch_dtype = torch.get_default_dtype() if not torch.cuda.is_available() else torch.bfloat16
+    device = "cuda" if torch.cuda.is_available() else "cpu"
     
     pipe = DiffusionPipeline.from_pretrained(
         model_path,
@@ -60,7 +69,7 @@ def load_model(model_path, ckpt_path=None, use_lora=True):
 
 model_id = "Qwen/Qwen-Image"
 lora_ckpt_path = "CIawevy/QwenImage-TextPecker-SQPA"
-device = "cuda"
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
 negative_prompt = " "
 aspect_ratios = {
@@ -87,4 +96,15 @@ image = pipe(
 
 image.save("TextPecker_qwen_demo.png")
 print("img has been saved to: TextPecker_qwen_demo.png")
+```
+
+## Citation
+
+```bibtex
+@article{zhu2026TextPecker,
+  title   = {TextPecker: Rewarding Structural Anomaly Quantification for Enhancing Visual Text Rendering},
+  author  = {Zhu, Hanshen and Liu, Yuliang and Wu, Xuecheng and Wang, An-Lan and Feng, Hao and Yang, Dingkang and Feng, Chao and Huang, Can and Tang, Jingqun and Bai, Xiang},
+  journal = {arXiv preprint arXiv:2602.20903},
+  year    = {2026}
+}
 ```