DisenStudio: Customized Multi-subject Text-to-Video Generation with Disentangled Spatial Control

Hong Chen, Xin Wang, Yipeng Zhang, Yuwei Zhou, Zeyang Zhang, Siao Tang, Wenwu Zhu

Abstract: Generating customized content in videos has received increasing attention recently. However, existing works primarily focus on customized text-to-video generation for single subject, suffering from subject-missing and attribute-binding problems when being applied to multiple subjects. Furthermore, existing models struggle to assign the desired actions to the corresponding subjects (action-binding problem), failing to achieve satisfactory multi-subject generation performance. To tackle the problems, in this paper, we propose DisenStudio, a novel framework that can generate text-guided videos for customized multiple subjects, given few images for each subject. Specifically, DisenStudio enhances a pretrained diffusion-based text-to-video model with our proposed spatial-disentangled cross-attention mechanism to associate each subject with the desired action. Then the pretrained model is customized for the multiple subjects with the proposed motion-preserved disentangled finetuning, which involves three tuning strategies: multi-subject co-occurrence tuning, masked single-subject tuning, and multi-subject motion-preserved tuning. The first two strategies guarantee the subject occurrence and preserve their visual attributes, and the third strategy helps the model to maintain the temporal motion-generation ability when finetuning on static images. We conduct extensive experiments to demonstrate that our proposed DisenStudio significantly outperforms existing methods in various metrics. Additionally, we show that DisenStudio can be used as a powerful tool for various controllable generation applications.

Primary Subject Area: [Generation] Generative Multimedia

Secondary Subject Area: [Experience] Multimedia Applications

Relevance To Conference: This work belongs to the area of Generative Multimedia, which focuses on generating videos from text prompts. The techniques proposed in this paper help better finetune the multimodal generative models to create higher-quality video content. Additionally, we involve the image finetuning for video generation models, therefore, our work is related to text, image, and video, a total of 3 modalities. In sum, this work is highly related to multimedia and multimodal processing.

Submission Number: 1342