StoryMovieScript: Visual Stories Grounded in Movie Scripts with Aligned Dialogue

StoryMovieScript is a multimodal dataset combining sequences of images with aligned screenplay dialogue and actions. The dataset contains 1,494 training samples and 263 test samples, with each story averaging around 13 images. It was created by author daniel3303 and last updated on Hugging Face in December 2025.

Use Cases

• Train models for visual story generation based on aligned screenplay dialogue and actions.
• Develop sequence-to-sequence models for image captioning or scene description based on the visual entity analysis (chain_of_thought).
• Benchmark multimodal alignment models using the paired image sequences and text from movie scripts.

Strengths

• Contains 1,494 training samples, providing a substantial base for model development.
• Each story includes 5-22 images, averaging around 13 frames, offering sequential visual context.
• Data is explicitly aligned, pairing image sequences with screenplay dialogue and actions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific movie scripts used as its source.

Provenance

Source

huggingface

Collection Method

Likely gathered by aligning publicly available movie scripts with corresponding image sequences.

Time Range

null

Freshness

Last updated 2025-12-24 19:38:36; freshness should be verified.

Geography

null