Abstract: In this work, we study the problem of Text-to-Image In-Context Learning (T2I-ICL). While Unified Multimodal LLMs (MLLMs) have advanced rapidly in recent years, they struggle with contextual reasoning in T2I-ICL scenarios. To address this limitation, we propose a novel framework that incorporates a thought process called ImageGen-CoT prior to image generation. To avoid generating unstructured ineffective reasoning steps, we develop an automatic pipeline to curate a high-quality ImageGen-CoT dataset. We then fine-tune MLLMs using this dataset to enhance their contextual reasoning capabilities. To further enhance performance, we explore test-time scale-up strategies and propose a novel hybrid scaling approach. This approach first generates multiple ImageGen-CoT chains and then produces multiple images for each chain via sampling. Extensive experiments demonstrate the effectiveness of our proposed method. Notably, fine-tuning with the ImageGen-CoT dataset leads to a substantial 80% performance gain for SEED-X on T2I-ICL tasks.

• 🎯 Chain-of-Thought Prompting: We propose a novel framework that incorporates a thought process called ImageGen-CoT prior to image generation in T2I-ICL tasks.

• [07/08/25] Our paper is accepted by ICCV2025
• [07/08/25] Our dataset is available on 🤗huggingface!

• Step 1: Set Up Environment
• Step 2: Download Dataset
• Step 3: Evaluation
• Step 4: Citation

1. Clone this repository and required models

   # Clone main repository
   git clone https://github.com/CurryX-001/ImageGen-CoT
   cd ImageGen-CoT/models
   
   # Clone and setup LLaVA
   git clone https://github.com/yzeng58/LLaVA 
   mv LLaVA llava
   
   # Clone and setup SEED-X (requires git-lfs)
   git lfs install
   git clone https://huggingface.co/spaces/tttoaster/SEED-X-17B
   mv SEED-X-17B SEED_X

2. Install Packages

   Linux

   # create the environment for llava to work 
   conda create -n llava python=3.10.13
   conda activate llava
   pip install --upgrade pip  # enable PEP 660 support
   pip install git+https://github.com/yzeng58/LLaVA/@a61aae093656922fe16ec2152b031dd1de72fe92
   pip install -r conda_env/llava_requirements.txt
   
   # create the environment for seed-x to work 
   conda env create -f conda_env/seedx_environment.yml

1. Download the CoBSAT dataset.

   wget "https://huggingface.co/datasets/yzeng58/CoBSAT/resolve/main/datasets.zip"

2. Extract the datasets.zip file using unzip datasets.zip and move the datasets folder into your ImageGen-CoT directory.

Please refer to the evaluation scripts in the scripts directory:

• scripts/baseline.sh: For running baseline
• scripts/evaluate.sh: For running the main evaluations

If you find this work helpful, please cite our paper:

@article{liao2025imagegen,
  title={Imagegen-cot: Enhancing text-to-image in-context learning with chain-of-thought reasoning},
  author={Liao, Jiaqi and Yang, Zhengyuan and Li, Linjie and Li, Dianqi and Lin, Kevin and Cheng, Yu and Wang, Lijuan},
  journal={arXiv preprint arXiv:2503.19312},
  year={2025}
}

This codebase is based on CoBSAT. We thank them for their great work!