Poster100K Dataset

A comprehensive dataset containing 93K+ movie and TV show posters with detailed captions and text region annotations for multimodal learning and poster generation tasks.

Dataset Structure

• image: Poster image in JPG/JPEG/PNG format
• caption: Detailed textual description generated by Gemini-2.5-flash-preview-04-17
• mask_regions: Text region coordinates (bounding boxes) in JSON format
• file_name: Original filename
• folder_path: Normalized relative folder path (batch-agnostic)
• batch_name: Source batch identifier (batch1, batch2, batch3)
• normalized_path: Standardized relative path without batch-specific prefixes

Usage

from datasets import load_dataset

dataset = load_dataset("PosterCraft/Poster100K")
print(dataset)

# Access a sample
sample = dataset['train'][0]
print(sample['caption'])
print("Text regions:", sample['mask_regions'])
print("Batch:", sample['batch_name'])
print("Normalized path:", sample['normalized_path'])
sample['image'].show()

Features

• High-quality posters: Movie and TV show posters from various genres and eras
• Detailed captions: Comprehensive descriptions including layout, typography, colors, and themes
• Text region masks: Precise bounding box coordinates for text elements in posters
• Multimodal ready: Perfect for training vision-language models

Applications

• Fine-tuning text-to-image models for poster generation or text rendering
• Training multimodal models for poster understanding and layout analysis
• Text region detection and OCR tasks
• Visual question answering on poster content
• Style transfer and poster design research

Dataset Statistics

• Total samples: 93,000+
• Image formats: JPG, JPEG, PNG
• Caption language: English
• Text regions: Bounding box coordinates in [x1, y1, x2, y2] format
• Average caption length: ~200-500 words
• Batches: 3 batches with normalized paths

License

This dataset is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).

Disclaimer

This dataset contains materials protected by third-party copyrights. The inclusion of these materials is based on the 'fair use' principle under copyright law and is intended solely for non-commercial scientific research purposes. We do not own the copyrights to these materials, and users are responsible for ensuring that their use complies with relevant laws and regulations. The publishers of this project assume no responsibility for any copyright disputes arising from the use of this dataset.

Citation

If you use this dataset in your research, please cite:

@article{chen2025postercraft,
  title={PosterCraft: Rethinking High-Quality Aesthetic Poster Generation in a Unified Framework},
  author={Chen, Sixiang and Lai, Jianyu and Gao, Jialin and Ye, Tian and Chen, Haoyu and Shi, Hengyu and Shao, Shitong and Lin, Yunlong and Fei, Song and Xing, Zhaohu and Jin, Yeying and Luo, Junfeng and Wei, Xiaoming and Zhu, Lei},
  journal={arXiv preprint arXiv:2506.10741},
  year={2025}
}

Contact

For questions or issues regarding this dataset, please open an issue on the Hugging Face dataset page.