Improve model card: Add pipeline tag, library name, links, and usage examples

README.md

@@ -1,7 +1,68 @@
 ---
 license: cc-by-nc-4.0
+pipeline_tag: image-to-image
+library_name: diffusers
 ---
 
-Models for [this repo](https://github.com/zengxianyu/ppd-examples) and [this paper](https://arxiv.org/abs/2512.05106) arxiv.org/abs/2512.05106
+# NeuralRemaster: Phase-Preserving Diffusion for Structure-Aligned Generation
 
-The 4-step lora checkpoints for Wan2.2-14b are converted from [Wan2.2-Lightning](https://huggingface.co/lightx2v/)
+This repository contains models for **NeuralRemaster**, also known as **Phase-Preserving Diffusion ($\phi$-PD)**, presented in the paper [NeuralRemaster: Phase-Preserving Diffusion for Structure-Aligned Generation](https://arxiv.org/abs/2512.05106).
+
+**NeuralRemaster** introduces a novel, model-agnostic reformulation of the diffusion process that preserves input phase while randomizing magnitude. This approach enables structure-aligned generation without requiring architectural changes or additional parameters, making it particularly well-suited for tasks demanding geometric consistency, such as re-rendering, simulation enhancement, and various image-to-image and video-to-video translation tasks. It also proposes Frequency-Selective Structured (FSS) noise for continuous control over structural rigidity.
+
+- 📚 **Paper**: [NeuralRemaster: Phase-Preserving Diffusion for Structure-Aligned Generation](https://arxiv.org/abs/2512.05106)
+- 🌐 **Project Page**: [https://yuzeng-at-tri.github.io/ppd-page/](https://yuzeng-at-tri.github.io/ppd-page/)
+- 💻 **Code**: [https://github.com/zengxianyu/PPD-examples](https://github.com/zengxianyu/PPD-examples)
+
+The 4-step LoRA checkpoints for Wan2.2-14b mentioned in this repository are converted from [Wan2.2-Lightning](https://huggingface.co/lightx2v/).
+
+## Usage
+
+This repository provides example adaptations of SD1.5, FLUX.1-dev, and Wan2.2-14b with Phase-Preserving Diffusion.
+
+1.  **Install dependencies**:
+
+    ```bash
+    pip install -r requirements.txt
+    pip install git+https://github.com/zengxianyu/structured-noise
+    ```
+
+2.  **Download model weights**:
+    Download the model weights from [huggingface.co/zengxianyu/ppd/tree/main](https://huggingface.co/zengxianyu/ppd/tree/main) and place them in `models/ppd/`.
+
+3.  **Inference examples**:
+    Example input images can be found [here](https://huggingface.co/zengxianyu/ppd/tree/main).
+
+    **SD 1.5**:
+
+    ```bash
+    PYTHONPATH=. python examples/image_synthesis/sd_text_to_image_ppd.py --input_image dog.jpg --radius 15 --prompt "A high quality picture captured by a professional camera. Picture of a cute border collie" --output output.png
+    ```
+
+    **FLUX1.1-dev**:
+
+    ```bash
+    PYTHONPATH=. CUDA_VISIBLE_DEVICES=6 python examples/flux/model_inference/FLUX.1-dev_ppd.py --input_image test2.jpg --prompt "$(cat test2.txt)" --output output.png --radius 30
+    ```
+
+    **Wan2.2-14b**:
+
+    ```bash
+    PYTHONPATH=. CUDA_VISIBLE_DEVICES=1 python examples/wanvideo/model_inference/Wan2.2-I2V-A14B_ppd.py --input_image output.png --input_video test2.mp4 --prompt "$(cat test2.txt)" --radius 30 --output output.mp4
+    ```
+
+4.  For training, please refer to the original [DiffSynth-Studio repository](https://github.com/modelscope/DiffSynth-Studio).
+
+## Citation
+
+If you find this work useful, please cite the paper:
+
+```bibtex
+@article{zeng2025neuralremaster,
+  title   = {{NeuralRemaster}: Phase-Preserving Diffusion for Structure-Aligned Generation},
+  author  = {Zeng, Yu and Ochoa, Charles and Zhou, Mingyuan and Patel, Vishal M and
+             Guizilini, Vitor and McAllister, Rowan},
+  journal = {arXiv preprint arXiv:XXXX.XXXXX},
+  year    = {2025}
+}
+```