license: mit
size_categories:
- 1M<n<10M
extra_gated_prompt: >-
Important Note: By accessing, requesting, downloading, or using this
dataset, you acknowledge and agree that it may contain harmful, offensive,
disturbing, explicit, unsafe, or otherwise NSFW prompts that could cause
psychological or physiological discomfort, distress, or other adverse effects;
that such content is inappropriate for display in public settings and must not
be accessed, viewed, or used by minors; that the dataset is provided strictly
for non-commercial academic research related to safety evaluation, red
teaming, and security analysis; that you bear full responsibility for ensuring
compliance with all applicable laws, ethical standards, and institutional
requirements; and that the authors, contributors, and maintainers assume no
liability for any direct, indirect, incidental, or consequential damages,
harms, misuse, or legal issues arising from your access to or use of the
dataset.
extra_gated_fields:
Name: text
Date of birth: date_picker
Country: country
Affiliation: text
Affiliation email: text
Job title:
type: select
options:- Student
- Research Graduate
- AI researcher
- AI developer/engineer
- Reporter
- Other
Intended Use (Please describe your intended use of this dataset. Overly casual, vague, or insufficiently justified requests may be declined. Providing relevant background information will strengthen your request.):
type: text
geo: ip_location
I confirm that I have read and understood the Important Notes section of this document and I agree to comply with all terms and requirements stated therein: checkbox
DREAM: Scalable Red Teaming for Text-to-Image Generative Systems
This dataset contains red-teaming prompts generated by the DREAM framework, as presented in the IEEE S&P 2026 paper: "DREAM: Scalable Red Teaming for Text-to-Image Generative Systems via Distribution Modeling". These prompts are designed to evaluate and stress-test the safety mechanisms of Text-to-Image (T2I) generative systems.
⚠️ Disclaimer / Warning
Content Warning: This dataset contains prompts that may be offensive, disturbing, explicit, unsafe, or otherwise NSFW (Not Safe For Work). These prompts are intended solely for academic research purposes related to AI safety, red teaming, and security analysis.Usage Restriction: By accessing or using this dataset, you acknowledge that the content is inappropriate for public display and must not be accessed by minors. The authors assume no liability for any misuse or consequences arising from the use of this dataset.
Dataset Format
The dataset is organized according to the following fields:
| Field | Description |
|---|---|
| prompt | The generated red-teaming text prompt designed to elicit unsafe content from the target system. |
| category | The specific type of unsafe content the prompt targets. Typically sexual or violence. |
| t2i_model | The specific Text-to-Image model targeted during the red teaming process (e.g., sd15, esd, ca, uce, safegen or rece). |
| filter | The safety filter that was active during the red teaming process (e.g., safety_checker, nsfw_image_detector, nsfw_text_detector, keyword_gibberish_filter or none). |
Citation
If you use this dataset in your research, please cite our paper:
@inproceedings{li2026dream,
title={DREAM: Scalable Red Teaming for Text-to-Image Generative Systems via Distribution Modeling},
author={Li, Boheng and Wang, Junjie and Li, Yiming and Hu, Zhiyang and Qi, Leyi and Dong, Jianshuo and Wang, Run and Qiu, Han and Qin, Zhan and Zhang, Tianwei},
booktitle={2026 IEEE Symposium on Security and Privacy (SP)},
year={2026},
organization={IEEE}
}