Progress in the field of artificial intelligence seems to be getting faster and faster. Generative AI models for videos are among the best examples to appreciate this. In line with this, a new launch by a very popular company promises to make deepfake videos 10x better.
ByteDance, does that name ring a bell? We are talking about the parent company of TikTok. Many may not even know that ByteDance offers other products, but such popular apps as CapCut belong to the Chinese firm. ByteDance has also entered the AI segment with investment projections of $12 billion by 2025, Reuters reported last month. The company had already launched some AI-powered tools in China in 2024, but now it has arrived with its most ambitious AI product to date.
ByteDance’s OmniHuman-1 AI model promises 10x better deepfakes
The firm has announced OmniHuman-1, an AI model for generating videos. At this point, the idea doesn’t seem new. There are already generative tools for video from multiple companies. However, it seems that ByteDance has broken into the market in a big way by offering the—potentially—best and most realistic one to date.
OmniHuman-1 doesn’t use the classic system of writing a prompt to generate the video. The service does so based on inputs in the form of photos, being compatible with both selfies and full-body photos. It can even detect and work with cartoon images. Then, you can add audio and/or video so that the character in the photo moves realistically based on the inputs.
What is most striking about OmniHuman-1 is the tremendously realistic results it achieves. It seems that we will soon reach the point where AI-generated videos become indistinguishable from reality. Artificial intelligence could achieve one of the big goals that the tech industry has had for a long time: breaking the “uncanny valley” effect.
X/Twitter user AshutoshShrivastava shared a set of samples generated with OmniHuman-1:
If the outputs are “as shown,” the results are pretty impressive. Hopefully, there won’t be news about samples being manipulated to look better, as with earlier services. It’s noteworthy that OmniHuman-1 can work with non-human characters as well.
Of course, there is a potential downside to this. We can’t expect deepfakes to become 10x better without thinking about potential uses by malicious actors. However, it’s up to companies to figure out ways to accurately determine when media content is AI-generated and label it as such.
The AI industry will continue to advance by leaps and bounds
Meanwhile, the growth in AI-based developments shows no signs of slowing down. Google, one of the biggest names in the industry, plans to invest $75 billion in AI this year alone. The DeepSeek models could also receive significant improvements, although it remains to be seen whether it can overcome all the controversy that has been surrounding it.
