Problem
Research questions and friction points this paper is trying to address.
Advertisement Video Generation
Product Identity Preservation
Large-scale Dataset
Evaluation Benchmark
Multi-shot Storytelling
AdSpark: A Large-Scale Dataset and Benchmark for Product-Centric Advertisement Video Generation
🤖 AI Summary
This study addresses the absence of large-scale dedicated datasets and systematic evaluation frameworks in product-centric advertising video generation. To this end, we introduce AdSpark, a unified dataset comprising 300,000 samples, alongside AdSpark-Bench, a six-dimensional diagnostic benchmark. Methodologically, we employ a hybrid data collection strategy combining real-world and synthetic sources, providing structured creative annotations and an automated evaluation system encompassing multiple dimensions such as visual quality and product fidelity. Through model fine-tuning experiments, we validate the effectiveness of the proposed dataset and reveal critical bottlenecks in existing models regarding product identity consistency and multi-shot narrative generation. Ultimately, this work bridges the evaluation gap in this specific domain and establishes a new research standard for product-oriented advertising video synthesis.
📝 Abstract
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 \textbf{AdSpark}, a large-scale dataset and benchmark for product-centric advertisement video generation, based on data from a major e-commerce platform. \textit{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 \textit{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.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Advertisement Video Generation
Large-Scale Dataset
Evaluation Benchmark
Product Identity Preservation
Multi-shot Storytelling
🔎 Similar Papers
No similar papers found.
