A Comprehensive Ecosystem for Open-Domain Customized Video Generation

2026-06-10

AI Summary

This work addresses the challenge of open-domain personalized video generation, which is hindered by the scarcity of large-scale, identity-annotated multimodal data. To this end, the authors introduce PexelsCustom-1M, a dataset comprising over one million identity–text–video triplets, and propose CustoMDiT, a parameter-efficient framework that adapts a pretrained multimodal diffusion Transformer into a personalized video generator with only 8% additional learnable parameters. By integrating cross-dataset knowledge and aligning video, text, and identity representations, the method achieves state-of-the-art performance on OpenCustom, a newly established benchmark encompassing more than 1,000 identity categories. The project fully releases the dataset, training pipeline, benchmark, and code to advance research in open-domain customized video generation.

Abstract

Recent progress in video generation has shown impressive visual synthesis capabilities. However, open-domain customized video generation remains limited by the lack of large-scale, annotated datasets capturing diverse identity-specific attributes. To address this, we introduce PexelsCustom-1M, the first publicly available million-scale dataset for identity-preserving video generation, containing one million curated <identity, text, video> triplets across 8,000+ categories. Leveraging this, we propose CustoMDiT, a parameter-efficient framework that adapts a pretrained multimodal Diffusion Transformer into a customized video generator with only 8% additional learnable parameters. Our method surpasses prior state-of-the-art. However, benchmarks such as DreamBooth cover only 100 classes, which is insufficient for real-world applications. To overcome this, we construct OpenCustom, a new benchmark with 1,000+ categories, created via cross-dataset knowledge fusion from ImageNet and MS-COCO. Extensive experiments confirm the advantages of both our dataset and model. We will open-source the entire ecosystem--including dataset, pipeline, benchmark, and implementations--to support further research.

Problem

• open-domain video generation
• customized video generation
• identity-preserving
• large-scale dataset
• benchmark

Innovation

• customized video generation
• PexelsCustom-1M
• CustoMDiT
• OpenCustom benchmark
• parameter-efficient adaptation