Unipic3-Consistency-Model

🌌 Unipic3-Consistency-Model

📖 Introduction

UniPic3-Consistency-Model is a few-step image editing and multi-image composition model based on Consistency Flow Matching (CM). The model learns a trajectory-consistent mapping from noisy latent states to clean images, enabling stable generation with strong structural consistency. It is distilled from UniPic-3 to support fast inference (≤8 steps) while preserving composition correctness.The model is especially suitable for scenarios requiring geometric alignment and semantic coherence, such as multi-image composition and human–object interaction (HOI).

📊 Benchmarks

🧠 Usage

1. Clone the Repository

git clone https://github.com/SkyworkAI/UniPic
cd UniPic-3

2. Set Up the Environment

conda create -n unipic python=3.10
conda activate unipic3
pip install -r requirements.txt

3.Batch Inference

transformer_path = "Skywork/Unipic3-Consistency-Model/ema_transformer"

python -m torch.distributed.launch --nproc_per_node=1 --master_port 29501 --use_env \
    qwen_image_edit_fast/batch_inference.py \
    --jsonl_path data/val.jsonl \
    --output_dir work_dirs/output \
    --distributed \
    --num_inference_steps 8 \
    --true_cfg_scale 4.0 \
    --transformer transformer_path \
    --skip_existing

📄 License

This model is released under the MIT License.

Citation

If you use Skywork UniPic 3.0 in your research, please cite:

@article{wei2026skywork,
  title={Skywork UniPic 3.0: Unified Multi-Image Composition via Sequence Modeling},
  author={Wei, Hongyang and Liu, Hongbo and Wang, Zidong and Peng, Yi and Xu, Baixin and Wu, Size and Zhang, Xuying and He, Xianglong and Liu, Zexiang and Wang, Peiyu and others},
  journal={arXiv preprint arXiv:2601.15664},
  year={2026}
}