metadata

license: apache-2.0

VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations for Synthetic Videos

Zongxia Li*, Xiyang Wu*, Yubin Qin, Guangyao Shi, Hongyang Du, Dinesh Manocha, Tianyi Zhou, Jordan Lee Boyd-Graber

[📖 Paper] [🤗 Dataset][🌍Website]

👀 About VideoHallu

With the recent success of video generation models such as Sora, Veo2, Kling, the visual quality of generated videos has reached new heights—making evaluation more challenging and pushing it beyond traditional metrics like frame consistency, resolution, and realism. However, we find that MLLMs struggle to detect abnormalities in generated videos, which is crucial for developing reliable automatic video evaluation methods.

We introduce VideoHallu, a curated dataset that includes videos generated by seven video generation models and a question-answer set to test MLLM's abilities to catch generated videos' abnormalities.

We also use GRPO to train Qwen-2.5-VL-7B on a subset of our dataset and show improvement on generated video understanding.

🔥 News

[2025/05/02] We release our datasets in huggingface🤗.

🔍 Dataset

To facilitate GRPO training, we also randomly sample 1,000 videos from PhysBench training data to first improve model' reasoning abilities in real-world videos, then train the model on part of our synthetic videos.

Our data spans the following categories:

Getting Started

# Download the dataset
pip install huggingface_hub

# Download data to your local dir
huggingface-cli download IntelligenceLab/VideoHallu --repo-type dataset --local-dir ./new_video_folders --local-dir-use-symlinks False

The Dawn of MLLMs in Synthetic Videos 🧠

🎬 Video: Quail Transforming into Rooster

Prompt (Sora): Generate a quail and a rooster celebrating New Year.

🎬 Video: Object Falling and Law of Physics

Prompt (Veo2): A feather and a heavy rock are released at the same height and begin to fall to the ground on Earth.

🎬 Video: Object Contact Abnormalities

Prompt (Sora): Generate a man drinking up a cup of wine.

🎬 Video: Breaking Process

Prompt (Sora): Generate the sequence showing a bullet being shot into a watermelon.

Acknowledgements

We sincerely appreciate the contributions of the open-source community. The related projects are as follows: R1-V , DeepSeek-R1 , Video-R1, Qwen-2.5-VL

Citations

If you find our work helpful for your research, please consider citing our work.

@article{feng2025video,
  title={Video-R1: Reinforcing Video Reasoning in MLLMs},
  author={Feng, Kaituo and Gong, Kaixiong and Li, Bohao and Guo, Zonghao and Wang, Yibing and Peng, Tianshuo and Wang, Benyou and Yue, Xiangyu},
  journal={arXiv preprint arXiv:2503.21776},
  year={2025}
}