RoPEMover: Depth-Aware Object Relocation via Positional Embeddings

Project page · Code · arXiv · Demo

İpek Öztaş, Duygu Ceylan, Aybars Buğra Aksoy, Ayşegül Dündar

This repository contains the RoPEMover LoRA (rank 16) for Qwen-Image-Edit-2511. RoPEMover moves an object in a single image by manipulating a depth-aware 3D extension of the diffusion transformer's rotary positional embeddings.

File	Description
ropemover_qwen_image_edit_2511_lora.safetensors	LoRA trained on 5k synthetic CLEVR pairs, then fine-tuned on real captured pairs

Usage

This LoRA requires the modified pipeline from the RoPEMover code repository. It does not work with the stock Qwen-Image-Edit pipeline.

git clone https://github.com/ipekoztas/RoPEMover.git && cd RoPEMover
pip install -r requirements.txt
python inference.py --image image.jpg --mask mask.png \
    --prompt "Move the mug to the right." --dx 300 --dy -40 --dz 0.1 --output out.png

The weights are downloaded from this repository automatically.

Citation

@article{oztas2026ropemover,
  title   = {RoPEMover: Depth-Aware Object Relocation via Positional Embeddings},
  author  = {Oztas, Ipek and Ceylan, Duygu and Aksoy, Aybars Bugra and Dundar, Aysegul},
  journal = {arXiv preprint arXiv:2606.27332},
  year    = {2026}
}

License

Apache 2.0. Use is also subject to the license of Qwen-Image-Edit-2511.