.jpg)
OmniHuman-1 is another generative artificial intelligence (AI) model that can easily generate fake videos. To generate a fake video, OmniHuman-1 only needs a picture and an audio sample. In a demo shared by the researchers, a video of NVIDIA CEO Jensen Huang singing was generated based on a picture of him while firing up the H200 GPU.
The model was trained using 19,000 hours of video content. In addition to generating new videos based on pictures, OmniHuman-1 can also edit videos with changes made to the original movements and original audio. An example is the video below where Taylor Swift's original singing audio is replaced with her now singing a song in Japanese from a separate audio source. Body and mouth movements are synchronized with the new audio quite well.
According to Bytedance, OmniHuman-1 can also be videos on various screen ratio sizes and is also better at producing realistic hand movements than previous AI video-generating models. The quality of the video generated depends on the quality of the pictures used as sources. To try OmniHuman-1, please go to the official GitHub page below.