Poster100K: 93,000+ Movie and TV Show Posters with Captions and Text Annotations

93,000+ movie and TV show posters are paired with detailed captions and text region annotations. The dataset was created by PosterCraft and last updated on 2025-09-17. It is designed for multimodal learning and poster generation tasks.

Use Cases

• Train multimodal models based on paired poster images and detailed captions.
• Develop text-to-image generation systems for creating posters based on textual descriptions.
• Improve OCR models for detecting and reading text in complex visual layouts using the annotated text region coordinates.
• Benchmark image understanding models on the task of describing visual and textual elements in promotional artwork.

Strengths

• Contains over 93,000 poster images, providing a substantial scale for training.
• Includes detailed captions generated by the Gemini-2.5-flash-preview-04-17 model.
• Provides text region coordinates (bounding boxes) for each poster, enabling precise localization tasks.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect temporal or source bias inherent to the collection of movie and TV show posters.

Provenance

Source

PosterCraft

Collection Method

Likely scraped or collected from various sources; includes generated captions and annotations.

Freshness

Last updated 2025-09-17 08:02:22; freshness should be verified.