ByteDance Introduces OmniHuman-1: An AI System for Creating Lifelike Videos from a Single Photo
ByteDance, the parent company of TikTok, has launched OmniHuman-1, an advanced AI system capable of generating incredibly lifelike videos of individuals performing various actions, such as speaking, gesturing, singing, and playing musical instruments—all from just a single photo.
According to a research paper published on the online open-access archive arXiv, “OmniHuman significantly outperforms existing methods, generating highly realistic human videos based on weak signal inputs, particularly in terms of audio.” The tool accommodates image inputs of any aspect ratio, whether portraits, half-body, or full-body images, producing more lifelike and high-quality results across diverse scenarios.
On the OmniHuman-1 project page, researchers showcase a variety of sample videos that highlight the system’s impressive capabilities. These examples feature dynamic hand and body movements from multiple angles, as well as animated representations of characters, animals, and even historical figures brought back to life.