DreamVideo-Omni: Omni-Motion Controlled Multi-Subject Video Customization with Latent Identity Reinforcement Learning
2026-03-12 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors developed DreamVideo-Omni, a new video generation system that can accurately control multiple people and their detailed movements in videos. They use a two-step training approach where the first step teaches the model about appearances, movements, and camera angles, while the second step helps the model keep each person's identity clear. To manage complex scenes with several subjects, they introduce special embeddings that link movements to specific people. Their system shows better ability to create high-quality videos where both who is in the video and how they move can be precisely controlled.
diffusion modelsvideo synthesismulti-subject identitymotion controlpositional embeddinglatent spaceidentity preservationmotion injectionvideo diffusion backbonereward learning
Authors
Yujie Wei, Xinyu Liu, Shiwei Zhang, Hangjie Yuan, Jinbo Xing, Zhekai Chen, Xiang Wang, Haonan Qiu, Rui Zhao, Yutong Feng, Ruihang Chu, Yingya Zhang, Yike Guo, Xihui Liu, Hongming Shan
Abstract
While large-scale diffusion models have revolutionized video synthesis, achieving precise control over both multi-subject identity and multi-granularity motion remains a significant challenge. Recent attempts to bridge this gap often suffer from limited motion granularity, control ambiguity, and identity degradation, leading to suboptimal performance on identity preservation and motion control. In this work, we present DreamVideo-Omni, a unified framework enabling harmonious multi-subject customization with omni-motion control via a progressive two-stage training paradigm. In the first stage, we integrate comprehensive control signals for joint training, encompassing subject appearances, global motion, local dynamics, and camera movements. To ensure robust and precise controllability, we introduce a condition-aware 3D rotary positional embedding to coordinate heterogeneous inputs and a hierarchical motion injection strategy to enhance global motion guidance. Furthermore, to resolve multi-subject ambiguity, we introduce group and role embeddings to explicitly anchor motion signals to specific identities, effectively disentangling complex scenes into independent controllable instances. In the second stage, to mitigate identity degradation, we design a latent identity reward feedback learning paradigm by training a latent identity reward model upon a pretrained video diffusion backbone. This provides motion-aware identity rewards in the latent space, prioritizing identity preservation aligned with human preferences. Supported by our curated large-scale dataset and the comprehensive DreamOmni Bench for multi-subject and omni-motion control evaluation, DreamVideo-Omni demonstrates superior performance in generating high-quality videos with precise controllability.