DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Models

DiffThinker introduces a novel Generative Multimodal Reasoning paradigm, establishing a diffusion-based reasoning framework. It reformulates multimodal reasoning as a native generative image-to-image task, achieving superior logical consistency and spatial precision in vision-centric tasks compared to traditional text-centric Multimodal Large Language Models (MLLMs).

Features

DiffThinker exhibits four core properties in its approach to vision-centric reasoning:

• Efficiency: Streamlined reasoning process.
• Controllability: Precise spatial and logical generation.
• Native Parallelism: Advantageous for complex reasoning steps.
• Collaboration: Works effectively across multiple domains (sequential planning, combinatorial optimization, constraint satisfaction, and spatial configuration).

Quick Start

To get started with DiffThinker, clone the official repository and install the necessary dependencies:

git clone https://github.com/lcqysl/DiffThinker.git
cd DiffThinker/DiffSynth-Studio
pip install -e .
pip install gymnasium

# (Optional) Install vLLM for OCR tasks
# we recommend installing it in a SEPARATE environment to avoid conflicts.
# pip install vllm

Inference & Evaluation

The test datasets used in our experiments are provided within each task's directory. We recommend using the same data to ensure the reproducibility of our results and to facilitate comparison with other models. If you wish to generate your own test data, please refer to the gen.txt file in each task directory.

cd Maze

# 1. Inference and Parsing
bash eval/gen_and_parse.sh

# 2. Evaluation
bash eval/eval_path.sh

# 3. Individual Inference
python ../DiffSynth-Studio/add/infer/infer.py
python ../DiffSynth-Studio/add/infer/infer_with_middle.py

Citation

@article{he2024diffthinker,
  title={DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Models},
  author={He, Zefeng and Qu, Xiaoye and Li, Yafu and Zhu, Tong and Huang, Siyuan and Cheng, Yu},
  journal={arXiv preprint arXiv:2512.24165},
  year={2024}
}