ByteDance Unveils AI Video Model Capable of Creating Lifelike Full-Body Animations

OmniHuman uses advanced training to generate realistic videos from a single photo, raising both excitement and ethical concerns.

OmniHuman, ByteDance's new AI model, can create realistic full-body videos of people talking, singing, and moving using just a single photo and audio input.
The system outperforms previous methods by generating natural gestures, synchronized speech, and detailed body movements, showcasing its versatility across various scenarios.
ByteDance trained OmniHuman on over 18,700 hours of video data using a multimodal approach combining text, audio, and body poses to enhance realism and reduce data wastage.
While the technology demonstrates potential for entertainment, education, and digital avatars, experts warn about risks such as deepfake misuse in misinformation and fraud.
OmniHuman is currently a research project with no public release date, but its viral demos, including a lifelike recreation of Albert Einstein, highlight its groundbreaking capabilities.