# Control Signals

DisPose is a method for controlling human image animation. It enhances video generation quality through motion field guidance and keypoint correspondence. This technology can generate videos from reference images and driving videos while maintaining consistency in motion alignment and identity information. DisPose generates dense motion fields from sparse motion fields and reference images, providing region-level dense guidance, while retaining generalization capabilities for sparse pose control. Moreover, it extracts diffusion features corresponding to pose keypoints from reference images and transfers these features to the target pose to provide unique identity information. The main advantages of DisPose include extracting more universal and effective control signals without needing additional dense inputs and improving video quality and consistency through a plug-and-play hybrid ControlNet without freezing existing model parameters.

MOFA-Video is a method that animates single images using various control signals. It employs sparse-to-dense (S2D) motion generation and flow-based motion adaptation techniques, effectively utilizing different types of control signals such as trajectories, keypoint sequences, and their combinations to animate single images. During training, sparse control signals are generated through sparse motion sampling, and different MOFA-Adapters are trained to animate videos using pre-trained SVDs. In inference, different MOFA-Adapters can be combined to jointly control frozen SVDs.