Refaçade: Editing Object with Given Reference Texture
Paper • 2512.04534 • Published
import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("fishze/Refacade", dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]Refaçade: Editing Object with Given Reference Texture
Youze Huang1,*
Penghui Ruan2,*
Bojia Zi3,*
Xianbiao Qi4,†
Jianan Wang5
Rong Xiao4
* Equal contribution. † Corresponding author.
🚀 Overview
Refaçade is a unified image–video retexturing model built upon the Wan2.1-based VACE framework. It edits the surface material of specified objects in a video using user-provided reference textures, while preserving the original geometry and background. We use Jigsaw Permutation to decouple structural information in the reference image and a Texture Remover to disentangle the original object’s appearance. This functionality enables users to explore diverse possibilities effectively.
🛠️ Installation
Our project is built upon Wan2.1-based VACE.
pip install -r requirements.txt
pip install wan@git+https://github.com/Wan-Video/Wan2.1
🏃♂️ Gradio Demo
You can use this gradio demo to retexture objects. Note that you don't need to compile the SAM2.
python app.py
📂 Download
First, download our checkpoints:
huggingface-cli download --resume-download fishze/Refacade --local-dir models
Next, download SAM2 sam2_hiera_large.pt and place it at:
sam2/SAM2-Video-Predictor/checkpoints/
We recommend to organize local directories as:
Refacade
├── ...
├── examples
├── models
│ ├── refacade
│ │ └── ...
│ ├── texture_remover
│ │ └── ...
│ └── vae
│ └── ...
├── sam2
└── ...
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support