Improve model card for Lego-Edit: Add metadata, links, abstract, and structure

README.md

@@ -1,3 +1,79 @@
----
-license: apache-2.0
----
+---
+license: cc-by-nc-4.0
+pipeline_tag: image-to-image
+library_name: transformers
+---
+
+# Lego-Edit: A General Image Editing Framework with Model-Level Bricks and MLLM Builder
+
+<p align="center">
+  <img src="https://github.com/xiaomi-research/lego-edit/raw/main/resources/lego_pic.png" alt="Lego-Edit" width="240"/>
+</p>
+
+Lego-Edit is an instruction-based image editing framework introduced in the paper [Lego-Edit: A General Image Editing Framework with Model-Level Bricks and MLLM Builder](https://huggingface.co/papers/2509.12883).
+
+-   📚 **Paper**: [Lego-Edit: A General Image Editing Framework with Model-Level Bricks and MLLM Builder](https://huggingface.co/papers/2509.12883)
+-   🌐 **Project Page**: https://xiaomi-research.github.io/lego-edit/
+-   💻 **Code / GitHub Repository**: https://github.com/xiaomi-research/lego-edit
+-   🚀 **Live Demo**: https://editdemo.ai.xiaomi.net/
+
+## Abstract
+Instruction-based image editing has garnered significant attention due to its direct interaction with users. However, real-world user instructions are immensely diverse, and existing methods often fail to generalize effectively to instructions outside their training domain, limiting their practical application. To address this, we propose Lego-Edit, which leverages the generalization capability of Multi-modal Large Language Model (MLLM) to organize a suite of model-level editing tools to tackle this challenge. Lego-Edit incorporates two key designs: (1) a model-level toolkit comprising diverse models efficiently trained on limited data and several image manipulation functions, enabling fine-grained composition of editing actions by the MLLM; and (2) a three-stage progressive reinforcement learning approach that uses feedback on unannotated, open-domain instructions to train the MLLM, equipping it with generalized reasoning capabilities for handling real-world instructions. Experiments demonstrate that Lego-Edit achieves state-of-the-art performance on GEdit-Bench and ImgBench. It exhibits robust reasoning capabilities for open-domain instructions and can utilize newly introduced editing tools without additional fine-tuning. The figure below showcases Lego-Edit's qualitative performance.
+
+<p align="center"><img src="https://github.com/xiaomi-research/lego-edit/raw/main/resources/case_pic.png" width="95%"></p>
+
+## ✨ Features
+
+Lego-Edit supports local editing, global editing, and multi-step editing as demonstrated in our tests, with corresponding results shown above. We discuss its feedback responsiveness and tool-extension capabilities in our paper.
+
+Additionally, Lego-Edit accepts mask inputs for precise editing region control. Example applications are provided here:
+
+<p align="center"><img src="https://github.com/xiaomi-research/lego-edit/raw/main/resources/maskcase1.png" width="95%"></p>
+
+<p align="center"><img src="https://github.com/xiaomi-research/lego-edit/raw/main/resources/maskcase2.png" width="95%"></p>
+
+You can try it and find more usages of this framework.
+
+## 🔥 Quick Start
+
+For detailed instructions on setting up the environment, downloading checkpoints, and running the Gradio WebUI, please refer to the [Quick Start section in the GitHub repository](https://github.com/xiaomi-research/lego-edit#--quick-start).
+
+## 💼 New Tools Integration
+
+Lego-Edit supports the integration of new tools. For guidance on how to add custom tools and make them usable by The Builder, please refer to the [New Tools Integration section in the GitHub repository](https://github.com/xiaomi-research/lego-edit#--new-tools-integration).
+
+## 📍 More Usages
+
+Some editing models are trained at a resolution of 768 via the ICEdit method. The corresponding trained [Single-Task-LoRA](https://huggingface.co/xiaomi-research/lego-edit/tree/main/loras) are provided. For independent usage of these LoRAs, refer to the usage instructions at [ICEdit](https://github.com/River-Zhang/ICEdit).
+
+<p align="center"><img src="https://github.com/xiaomi-research/lego-edit/raw/main/resources/lora_effect.png" width="95%"></p>
+
+## 📄 Disclaimer
+
+We open-source this project for academic research. The vast majority of images 
+used in this project are either generated or licensed. If you have any concerns, 
+please contact us, and we will promptly remove any inappropriate content. 
+Our code is released under the Apache 2.0 License, while our models are under 
+the CC BY-NC 4.0 License. Any models related to <a href="https://huggingface.co/black-forest-labs/FLUX.1-dev" target="_blank">FLUX.1-dev</a> 
+base model must adhere to the original licensing terms.
+<br><br>This research aims to advance the field of generative AI. Users are free to 
+create images using this tool, provided they comply with local laws and exercise 
+responsible usage. The developers are not liable for any misuse of the tool by users.
+
+## ✍️ Citation
+
+If you find this project useful for your research, please consider citing our paper:
+
+```bibtex
+@article{jia2025legoedit,
+  title   = {Lego-Edit: A General Image Editing Framework with Model-Level Bricks and MLLM Builder},
+  author  = {Qifei Jia and Yu Liu and Yajie Chai and Xintong Yao and Qiming Lu and Yasen Zhang and Runyu Shi and Ying Huang and Guoquan Zhang},
+  journal = {arXiv preprint arXiv:2509.12883},
+  year    = {2025},
+  url     = {https://arxiv.org/abs/2509.12883}
+}
+```
+
+## 🙏 Acknowledgments
+
+- Built on the [MiMo-VL](https://github.com/XiaomiMiMo/MiMo-VL), [ComfyUI](https://github.com/comfyanonymous/ComfyUI), [FLUX](https://github.com/black-forest-labs/flux), [ICEdit](https://github.com/River-Zhang/ICEdit), [EVF-SAM](https://github.com/hustvl/EVF-SAM) and [LaMa](https://github.com/advimman/lama)