---
language:
- en
---
# Thinking with Generated Images
<p align="center" width="100%">
<img src="https://raw.githubusercontent.com/GAIR-NLP/thinking-with-generated-images/main/assets/logo.png" alt="thinking-with-generated-images" height="450" style="width:85%; min-width: 100px; display: block; margin: auto;">
</p>

We introduce **Thinking with Generated Images**, where we enable a single LMM (Large Multimodal Model) to spontaneously generate and reason with intermediate visual thoughts via a native long-multimodal thought process.
<p align="center" width="100%">
<img src="https://raw.githubusercontent.com/GAIR-NLP/thinking-with-generated-images/main/assets/framework.png" alt="thinking-with-generated-images" height="450" style="width:85%; min-width: 100px; display: block; margin: auto;">
</p>

This model supports vision generation with intermediate visual subgoals.
<p align="center" width="100%">
<img src="https://raw.githubusercontent.com/GAIR-NLP/thinking-with-generated-images/main/assets/geneval.png" alt="thinking-with-generated-images" height="450" style="width:85%; min-width: 100px; display: block; margin: auto;">
</p>

Please refer to [our github repo](https://github.com/GAIR-NLP/thinking-with-generated-images) for more information!