License Notice
The license shown in the repository metadata is incorrect due to Hugging Face interface limitations.
This LoRA is a derivative work of FLUX.1-dev and is therefore subject to:
flux-1-dev-non-commercial-license
Please refer to the official FLUX repository for the full license terms.
Shine LoRA for Physically Plausible Image Composition
This repository provides the LoRA weights used in our paper:
Does Flux already know how to perform physically plausible image composition?
This LoRA is trained to improve physically plausible image composition when used with FLUX.1-dev.
Base Model
This LoRA must be used together with:
black-forest-labs/FLUX.1-dev
You must download the base model separately from its official source.
Training Data
This LoRA was trained using:
- Shine-DreamEditBench (a reformatted version of DreamEditBench)
Usage
Load the LoRA weights into FLUX.1-dev following standard LoRA loading procedures.
License
This LoRA is a derivative work of FLUX.1-dev.
Therefore, usage of this LoRA is subject to the license of the base model:
flux-1-dev-non-commercial-license
Please refer to the official FLUX repository for full license terms.
Citation
If you use this LoRA, please cite:
1@article{lu2025does,
2 title={Does flux already know how to perform physically plausible image composition?},
3 author={Lu, Shilin and Lian, Zhuming and Zhou, Zihan and Zhang, Shaocong and Zhao, Chen and Kong, Adams Wai-Kin},
4 journal={arXiv preprint arXiv:2509.21278},
5 year={2025}
6}
