ScottHan/PositionIC

PositionIC: Unified Position and Identity Consistency for Image Customization

Junjie Hu, Tianyang Han, Kai Ma, Jialin Gao, Song Yang
Xianhua He, Junfeng Luo, Xiaoming Wei, Wenqiang Zhang

---

🔥 News

• ✅ [2026.01.12] We have released our PositionIC model for FLUX on HuggingFace and github!
• ✅ [2025.07.18] Our paper is now available on arXiv.
• ⬜ Datasets and PositionIC-v2 model with enhanced generation capabilities are coming soon.

---

📖 Introduction

PositionIC is a unified framework for high-fidelity, spatially controllable multi-subject image customization. While recent methods excel in fidelity, fine-grained instance-level spatial control remains a challenge due to the entanglement of identity and layout.

To address this, we introduce:

1. BMPDS: The first automatic data-synthesis pipeline for position-annotated multi-subject datasets, providing crucial spatial supervision.
2. Lightweight Layout-Aware Diffusion: A framework integrating a novel visibility-aware attention mechanism that explicitly models spatial relationships via NeRF-inspired volumetric weight regulation.

Our experiments demonstrate that PositionIC achieves state-of-the-art performance, setting new records for spatial precision and identity consistency in multi-entity scenarios.

---

⚡️ Quick Start

🔧 Requirements and Installation

Follow these steps to set up your environment:

# 1. Create and activate a new conda environment
conda create -n PositionIC python=3.10 -y
conda activate PositionIC

# 2. Install PyTorch (adjust according to your CUDA version)
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121

# 3. Install project dependencies
pip install -r requirements.txt

📥 Checkpoints Download

You can download the .safetensors weights (e.g., dit_lora.safetensors) using huggingface-cli:

pip install huggingface_hub

# Replace [YOUR_REPO] with your actual Hugging Face repository path
repo_name="ScottHan/PositionIC"
local_dir="models/"$repo_name

huggingface-cli download $repo_name --local-dir $local_dir

---

✍️ Inference

To generate images with precise position and identity control, run the following command:

python inference_.py \
  --eval_json_path "path/to/your/val_config.json" \
  --dit_lora_path "models/PositionIC/dit_lora.safetensors" \
  --saved_dir "./res" \
  --width 1024 \
  --height 1024 \
  --ref_size 512 \
  --seed 3074 \
  --rope_type "uno" \
  --a 5

---

🙏 Acknowledgments

Our code is built upon the UNO framework. We sincerely thank the authors for their excellent work and open-source contributions.

---

🌟 Citation

If you find our work helpful for your research, please consider giving us a star ⭐ and citing our paper:

@article{hu2025positionic,
  title={PositionIC: Unified Position and Identity Consistency for Image Customization},
  author={Hu, Junjie and Han, Tianyang and Ma, Kai and Gao, Jialin and Yang, Song and He, Xianhua and Luo, Junfeng and Wei, Xiaoming and Zhang, Wenqiang},
  journal={arXiv preprint arXiv:2507.13861},
  year={2025}
}

---

📄 License

This project is licensed under the Apache-2.0 License.

---