Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces AdSpark, a large-scale dataset and benchmark designed to advance product-centric advertisement video generation by addressing the scarcity of relevant data and evaluation frameworks. AdSpark-300K comprises approximately 300K image-prompt-video triplets with detailed annotations for product identity, selling points, and creative plans, enabling models to learn product preservation and narrative storytelling. AdSpark-Bench, a diagnostic benchmark, evaluates generated videos across six dimensions, revealing current model limitations in product fidelity and multi-shot storytelling.
A 300K-sample dataset and diagnostic benchmark unlock the challenging task of generating product-preserving, narrative-driven advertisement videos.
Product-centric advertisement video generation aims to create promotional videos that preserve fine-grained product identity while presenting selling points through coherent multi-shot narratives. However, this emerging task remains underexplored due to the lack of large-scale advertisement-specific datasets and comprehensive evaluation frameworks. To address this gap, we introduce AdSpark, a large-scale dataset and benchmark for product-centric advertisement video generation, based on data from a major e-commerce platform. AdSpark-300K contains approximately 300K reference image--prompt--video triplets, comprising a real-world subset and a synthetic subset. Each sample provides structured advertisement annotations, including product identity annotations, selling-point descriptions, creative plans, and aligned audio scripts, enabling models to learn product preservation and advertisement-oriented visual storytelling. We further propose AdSpark-Bench, a diagnostic benchmark that evaluates generated advertisements across six dimensions, including visual quality, product fidelity, instruction adherence, temporal coherence, audio alignment, and advertisement effectiveness. Based on AdSpark-Bench, we evaluate representative models, revealing key challenges in product preservation, multi-shot storytelling, and selling-point visualization. Experiments with AdSpark-300K-finetuned models further validate the effectiveness of our dataset. AdSpark provides a unified dataset and benchmark for future research, and we will release the dataset upon acceptance.