VisualCloze: A Universal Image Generation Framework via Visual In-Context Learning

Graph200k is a large-scale dataset containing a wide range of distinct tasks of image generation.

If you find Graph200k is helpful, please consider to star ⭐ the Github Repo. Thanks!

📰 News

• [2025-5-15] 🤗🤗🤗 VisualCloze has been merged into the official pipelines of diffusers. For usage guidance, please refer to the Model Card.
• [2025-5-18] 🥳🥳🥳 We have released the LoRA weights supporting diffusers at LoRA Model Card 384 and LoRA Model Card 512.

🌠 Key Features

• Each image is annotated for five meta-tasks, including 1) conditional generation, 2) image restoration, 3) image editing, 4) IP preservation, and 5) style transfer.
• Using these tasks, we can also combine a wide range of complex tasks. For example, Style+Subject+Layout to Image as shown below.

🔥 Quick Start

Uasge

import datasets
grapth200k = datasets.load_dataset("lzyhha/test") # todo
train = grapth200k['train']
test = grapth200k['test']
# Reading depth map (PIL.Image) of the first image in the train set 
train[0]['depth'].save(f'depth.jpg')

Quality filtering

Graph200K is built based on Subjects200K, which provides quality scores. In our VisualCloze method, we filter out image pairs that exhibit low subject consistency, specifically those where objectConsistency <= 3.

Annotations

In each item of the dataset, there are annotations as follows. We leave the discussions about data construction in our paper.

Citation

If you find VisualCloze useful for your research and applications, please cite using this BibTeX:

@InProceedings{Li_2025_ICCV,
    author    = {Li, Zhong-Yu and Du, Ruoyi and Yan, Juncheng and Zhuo, Le and Li, Zhen and Gao, Peng and Ma, Zhanyu and Cheng, Ming-Ming},
    title     = {VisualCloze: A Universal Image Generation Framework via Visual In-Context Learning},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2025},
    pages     = {18969-18979}
}