MARBLE

This is the official implementation of MARBLE: Material Recomposition and Blending in CLIP-Space. Given an input image, MARBLE enables edits from material transfer, blending, to parametric control.

Installation

MARBLE material blocks are built from the code base of InstantStyle. Additional functions are added into the ip_adapter_instantstyle/ip_adapter.py but please cite their papers accordingly.

The current code base is tested on Python 3.9.7.

We will begin by cloning this repo:

git clone https://github.com/Stability-AI/marble.git

Then, install the latest the libraries with:

cd marble
pip install -r requirements.txt

Usage

After installation and downloading the models, you can use the two demos try_blend.ipynb for material blending and parametric_control.ipynb for material transfer + multi-attribute parametric control.

We also provide a gradio demo which can be run with python gradio_demo.py.

Using your own materials

For material transfer, you could add your images input_images/texture/.

ComfyUI extension

Custom nodes and an example workflow are provided for ComfyUI.

To install:

• Clone this repo into custom_nodes:

 $ cd ComfyUI/custom_nodes
 $ git clone https://github.com/Stability-AI/marble

• Install dependencies:

 $ cd marble
 $ pip install -r requirements.txt

• Restart ComfyUI

Citation

If you find MARBLE helpful in your research/applications, please cite using this BibTeX:

@article{cheng2024marble,
  title={MARBLE: Material Recomposition and Blending in CLIP-Space},
  author={Cheng, Ta-Ying and Sharma, Prafull and Boss, Mark and Jampani, Varun},
  journal={CVPR},
  year={2025}
}