EAVSD: E-commerce Advertising Video Storyboard Dataset for Multi-Image Generation

EAVSD is a large-scale dataset for subject-driven, narrative-planned multi-image generation, released as part of a CVPR 2026 paper. Each sample contains a reference product image and eight cinematic scene images depicting the same subject in diverse, narratively coherent settings. The dataset was created by author zjyao-PKU and was last updated on 2026-05-26.

Use Cases

• Train models for subject-driven image generation based on a clean product photo anchor.
• Develop algorithms for narrative-planned visual storyboarding based on the described scene structure.
• Research long-range visual consistency in multi-image sequences based on the provided cinematic scene sets.
• Benchmark models for e-commerce advertising content creation based on the product-centric narrative theme.

Strengths

• Dataset is described as large-scale, suggesting a substantial number of samples.
• Each sample contains a structured set of 9 images (1 reference + 8 scenes), providing a consistent data format.
• Dataset is associated with a CVPR 2026 publication, indicating a research-grade origin.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

zjyao-PKU on Hugging Face

Collection Method

Likely created for the research paper 'Narrative Weaver: Towards Controllable Long-Range Visual Consistency with Multi-Modal Conditioning' (CVPR 2026).

Freshness

Last updated 2026-05-26 03:14:26; freshness should be verified.