Add paper link, project page, and GitHub repository

README.md

@@ -1,21 +1,23 @@
 ---
 license: apache-2.0
-tags:
-  - video
-  - relighting
-  - inverse-rendering
-  - diffusion
-  - cosmos
 pipeline_tag: image-to-image
+tags:
+- video
+- relighting
+- inverse-rendering
+- diffusion
+- cosmos
 ---
 
 # World Inverse Renderer
 
-Video inverse rendering model based on NVIDIA Cosmos 7B video diffusion transformer, fine-tuned on [custom dataset](https://github.com/ShandaAI/AlayaRenderer?tab=readme-ov-file).
+[**Project Page**](https://alaya-studio.github.io/renderer) | [**Paper**](https://huggingface.co/papers/2604.02329) | [**GitHub**](https://github.com/ShandaAI/AlayaRenderer)
+
+This repository contains the **World Inverse Renderer** model presented in the paper [Generative World Renderer](https://huggingface.co/papers/2604.02329).
 
 ## Model Description
 
-This model performs **inverse rendering** on images and videos: given an input RGB frame, it estimates physically-based G-buffer maps:
+This model performs **inverse rendering** on images and videos. Based on the NVIDIA Cosmos 7B video diffusion transformer, it has been fine-tuned on a large-scale dataset of 4M frames extracted from AAA games. Given an input RGB frame or video, it estimates physically-based G-buffer maps:
 
 - **Basecolor** (albedo)
 - **Normal** (surface normals)
@@ -23,16 +25,18 @@ This model performs **inverse rendering** on images and videos: given an input R
 - **Roughness**
 - **Metallic**
 
-These G-buffers can then be used with a forward renderer to relight the scene under arbitrary environment lighting (HDRI maps).
+These G-buffers can subsequently be used with a forward renderer to relight or edit the scene under arbitrary lighting conditions.
 
 ## Architecture
 
-- Based on NVIDIA Cosmos 7B video diffusion transformer
-- Fine-tuned on [custom dataset](https://github.com/ShandaAI/AlayaRenderer?tab=readme-ov-file)
-- Supports both single-image and multi-frame video inverse rendering
+- Based on NVIDIA Cosmos 7B video diffusion transformer.
+- Fine-tuned on the AlayaRenderer dataset (4M+ frames from AAA games like Cyberpunk 2077 and Black Myth: Wukong).
+- Supports both single-image and multi-frame video inverse rendering.
 
 ## Usage
 
+For environment setup and full inference code, please refer to the [official GitHub repository](https://github.com/ShandaAI/AlayaRenderer).
+
 ```bash
 # Inverse rendering on images
 CUDA_HOME=$CONDA_PREFIX PYTHONPATH=$(pwd) python cosmos_predict1/diffusion/inference/inference_inverse_renderer.py \
@@ -52,3 +56,14 @@ CUDA_HOME=$CONDA_PREFIX PYTHONPATH=$(pwd) python cosmos_predict1/diffusion/infer
 - Python 3.10
 - NVIDIA GPU with >= 16GB VRAM (48GB+ recommended)
 - CUDA 12.0+
+
+## Citation
+
+```bibtex
+@article{huang2026generativeworldrenderer,
+    title={Generative World Renderer},
+    author={Zheng-Hui Huang and Zhixiang Wang and Jiaming Tan and Ruihan Yu and Yidan Zhang and Bo Zheng and Yu-Lun Liu and Yung-Yu Chuang and Kaipeng Zhang},
+    journal={arXiv preprint arXiv:2604.02329},
+    year={2026}
+}
+```