Omnihuman is a human video generation model developed by ByteDance Research, designed to generate highly realistic video of human figures driven by audio or motion signals. It addresses the specific challenge of producing natural, full-body human video with coherent lip sync, body motion, and expressive performance, making it particularly relevant for applications in digital avatars, synthetic presenters, and character animation driven by audio input.

The model is notable for its ability to handle a range of human body types, poses, and motion inputs while maintaining high visual fidelity and temporal consistency across frames. Omnihuman can generate video where a human figure's speech, facial expression, and body language are all driven by an audio signal, producing results where the spoken performance and physical presence of the generated figure feel coherent and natural together. This integrated approach to audio-driven human video generation represents a step beyond simpler lip-sync tools by incorporating whole-body dynamics into the generation process. The model was introduced as a research contribution demonstrating state-of-the-art capability in the human video generation domain.

For creators working with synthetic presenters, digital doubles, or AI-generated human performances, models like Omnihuman expand the range of content that can be produced without live actors. As this category of tool matures, the combination of audio-driven human generation with consistent visual identity across multiple outputs will become increasingly relevant to content creation workflows.