Datasets:
task_categories:
- any-to-any
language:
- en
tags:
- multimodal
- interleaved-generation
pretty_name: IntReal-100K
size_categories:
- 100K<n<1M
📄 Paper · 🌐 Project Page · 🤖 Models · 🤗 Dataset · 🏆 Leaderboard · 💻 Code · 🎬 Video
IntReal-100K
IntReal-100K is a real-world dataset for multimodal interleaved generation, covering four categories: storytelling, knowledge, planning, and tutorials. We provide task instructions and curated interleaved text-image sequences. The IntReal evaluation split additionally includes fine-grained evaluation rubrics.
Data Sources
| Category | Sources |
|---|---|
| Storytelling | StoryStream: Curious George, Rabbids Invasion, and The Land Before Time |
| Knowledge | Wikipedia |
| Planning | Wikitravel and Wikivoyage |
| Tutorials | 下厨房 (Xiachufang) |
We thank the original creators and dataset contributors for their work.
License and Attribution
Original text and images remain subject to the licenses and rights of their respective sources. IntReal annotations do not replace or override those terms, and inclusion in this dataset does not grant additional rights to the underlying content.
For questions about attribution or content removal, please contact shaohan.yu@smail.nju.edu.cn.
Citation
@misc{yu2026intrealrealworlddatasetbenchmark,
title={IntReal: A Real-World Dataset and Benchmark for Multimodal Interleaved Generation},
author={Shaohan Yu and Jie Qin and Wende Tan and Haiwen Diao and Limeng Qiao and Yueming Lyu and Chenyang Si},
year={2026},
eprint={2610.38425},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2610.38425},
}