HunyuanCustom

HunyuanCustom is a multimodal-driven architecture for customized video generation developed by Tencent, built upon the HunyuanVideo framework. It emphasizes subject consistency while supporting flexible user-defined conditions across multiple input modalities, including text, images, audio, and video. The framework introduces an image-text fusion module based on LLaVA for enhanced multimodal understanding, an image ID enhancement module that uses temporal concatenation to reinforce identity features across frames, an AudioNet module for hierarchical audio condition injection via spatial cross-attention, and a video-driven injection module that integrates latent-compressed conditional video through a patchify-based feature-alignment network. Extensive experiments demonstrate significant improvements over state-of-the-art open- and closed-source methods in terms of identity consistency, realism, and text-video alignment, and the system shows robustness in downstream tasks such as audio- and video-driven customized video generation.