ByteDance launches OmniHuman-1 AI video generator that creates ultra-realistic videos from any photo
Chinese company ByteDance, the parent company of TikTok, has announced a new artificial intelligence framework called OmniHuman-1. This tool can convert a single image and audio recording into a realistic video with accurate facial and body movements.
Here's What We Know
OmniHuman-1 uses data-mixing technology, which allows it to create more realistic videos compared to other models. ByteDance researchers said that the development of OmniHuman-1 involved training on 19,000 hours of footage, which significantly improved the quality of the generated videos.
The tool supports images of any aspect ratio, be it portraits or full-length images. OmniHuman-1 can generate realistic hand movements, facial expressions and body gestures, making it ideal for virtual reality image generation, educational videos and other purposes. An example of such generation with a photo of Albert Einstein can be seen below.
OmniHuman-1 is still in the research phase, but its potential is already generating a lot of interest. When it becomes available to the general public, it will compete with other tools such as Sora and Veo from OpenAI and Google respectively.
Source: omnihuman
