---
license: apache-2.0
pipeline_tag: text-to-image
library_name: diffusers
---

# InstanceAssemble
> Official implementation of "InstanceAssemble: Layout-Aware Image Generation via Instance Assembling Attention" (NeurIPS 2025).

<p align="center">
  <img src="https://github.com/FireRedTeam/InstanceAssemble/raw/main/fig/teaser.jpg" alt="Teaser of InstanceAssemble" width="800">
</p>

This repository contains the model used in [InstanceAssemble: Layout-Aware Image Generation via Instance Assembling Attention](https://arxiv.org/abs/2509.16691).

## Introduction

InstanceAssemble is a lightweight framework for Layout-to-Image generation that enables precise spatial control. We also introduce DenseLayout and Layout Grounding Score (LGS) for rigorous evaluation, where InstanceAssemble achieves state-of-the-art performance on both sparse and dense layouts.

For the official code and more details, please refer to the GitHub repository: [https://github.com/FireRedTeam/InstanceAssemble](https://github.com/FireRedTeam/InstanceAssemble).

## Usage

### Inference
```bash
# sd3 based
python inference.py --model_type sd3 --input_json ./demo/bigchair.json
# flux based
python inference.py --model_type fluxdev --input_json ./demo/bigchair.json
python inference.py --model_type fluxschnell --input_json ./demo/bigchair.json
```

### Streamlit demo
```bash
streamlit run demo.py
```

## Citation

```
@article{xiang2025instanceassemble,
      title={InstanceAssemble: Layout-Aware Image Generation via Instance Assembling Attention}, 
      author={Qiang Xiang and Shuang Sun and Binglei Li and Dejia Song and Huaxia Li and Nemo Chen and Xu Tang and Yao Hu and Junping Zhang},
      journal={arXiv preprint arXiv:2509.16691},
      year={2025},
}
```