EAVSD is an e-commerce advertising video storyboard dataset released in 2026 by a team from Peking University. It aims to support subject-oriented multi-image generation and narrative planning tasks. Related research papers include... Narrative Weaver: Towards Controllable Long-Range Visual Consistency with Multi-Modal ConditioningIt is widely used in subject-oriented multi-image generation and narrative planning tasks, with a core focus on the generation of storyboards for e-commerce advertising videos and research on controllable long-range visual consistency. This dataset contains 50,538 product samples and a total of 401,351 scene images, covering 8 anonymized e-commerce product categories. Each sample includes one reference image, 8 English scene prompts, and the corresponding generated scene image. The reference image was extracted and cleaned from the original product list by the Qwen-VL model, and the scene prompts were planned and expanded into cinematic English descriptions by a large language model. The final scene images were generated using Novita AI's image editing API, and all scene images are model-synthesized data.

Data Fields

• id: Unique identifier for the asset (e.g., category_a_000123)
• category: The product category to which it belongs (category_a ~ category_h)
• Reference: File name of the referenced product image
• n_scenes_generated / n_scenes_total: The actual number of scenes generated versus the total planned number (fixed at 8).
• scene_prompts: A list of English prompts for the corresponding scene (preserving the generated template prefix).
• scene_filenames / scene_errors: List of scene graph filenames and corresponding error messages for failed scenes.

Citation

@article{yao2026narrative,
title   = {Narrative Weaver: Towards Controllable Long-Range Visual Consistency with Multi-Modal Conditioning},
author  = {Yao, Zhengjian and Li, Yongzhi and Gao, Xinyuan and Chen, Quan and Jiang, Peng and Lu, Yanye},
journal = {arXiv preprint arXiv:2603.06688},
year    = {2026}
}