ByteDance Unveils OmniHuman-1: Ultra-Realistic AI Video Technology

ByteDance's OmniHuman-1 is advancing multimodal AI, creating hyper-realistic videos of people speaking, singing, and moving with unprecedented smoothness.

ByteDance Unveils New AI Model

ByteDance, the tech giant behind TikTok, has made a significant entry into the artificial intelligence (AI) arena with the unveiling of its latest creation, OmniHuman-1. This technological breakthrough, impressively realistic, has the ability to animate images and sound recordings, showcasing China’s innovation prowess in the AI sector.

OmniHuman-1: A True Technological Marvel

OmniHuman-1 is a multimodal model that can generate lifelike videos of people talking, singing, and moving with a quality “far superior to existing methods of audio-conditioned human video generation,” according to ByteDance developers. It’s crucial to note that “the AI-generated images, videos, and audios of real people are often referred to as deepfakes”, a technology used both in fraudulent activities and for more playful purposes.

The Rise of AI in China

ByteDance demonstrates the growing fascination with AI in China. Evidence of this is the company’s Doubao app, currently the top player in the consumer sector. However, OmniHuman-1 is not yet available to the public, though snippets have already garnered substantial online interest.

Among these demonstrations, a 23-second clip features a realistic Albert Einstein delivering a speech. According to Kyle Wiggers of TechCrunch, the quality is “intentionally good” and “probably one of the most realistic deepfakes to date”.

This significant advancement in AI highlights the progress made by Chinese developers, despite Washington’s attempts to curb these developments. The launch of OmniHuman-1 follows the release of OpenAI’s video generation tool Sora to ChatGPT Plus and Pro users in December last year.