Chinese tech giant ByteDance, the parent company of TikTok, has quietly introduced an advanced artificial intelligence model capable of generating realistic human videos from a single still image. The unveiling of the OmniHuman-1 model puts ByteDance ahead of U.S. competition in AI video generation but raises national security concerns about the potential for deepfake abuse.
According to a research paper published by ByteDance, OmniHuman-1 can create high-fidelity videos of people talking and moving naturally with unprecedented accuracy and personalization. The technology, which was trained on over 18,700 hours of human video footage, allows users to produce AI-generated content that appears authentic and can evade traditional deepfake detection tools.
AI experts warn that the model could be exploited for disinformation, election interference, and fraudulent activities if made widely available. Henry Ajder, a leading expert on generative AI, told ABC News that the ability to generate a convincing video from just one image makes it easier to create deceptive content.
“Previously, you might have needed thousands of images to train these models, but now, with just one picture, you can fabricate a completely realistic video,” Ajder said.
The release of OmniHuman-1 comes amid heightened scrutiny of ByteDance by U.S. lawmakers, who have raised concerns over the company’s ties to the Chinese government. Experts fear that the AI tool could be weaponized for state-backed propaganda campaigns or used to create non-consensual deepfake content.
ByteDance has not publicly commented on OmniHuman-1, but in a statement to Forbes, the company claimed that any future deployment of the tool would include safeguards against misinformation and abuse. The company also pointed to TikTok’s existing policies, which require AI-generated content to be labeled.
Among the AI-generated videos showcased in ByteDance’s research paper were clips of Albert Einstein appearing to deliver a lecture, TED Talk speakers animated from still images, and musicians playing instruments while singing. The technology enables seamless lip-syncing and expressive movements, setting a new benchmark for AI-generated video.
John Cohen, a former intelligence official at the U.S. Department of Homeland Security, warned that AI-generated content could be exploited by foreign adversaries to manipulate public opinion.
“The ability to generate hyper-realistic deepfake videos could significantly expand the spread of disinformation, particularly in the lead-up to the 2024 elections,” Cohen said.
Recent reports indicate that AI-generated content has already been used in political campaigns worldwide. In Moldova, deepfake videos falsely depicted the country’s president endorsing a pro-Russian party, while in Bangladesh, AI-generated images were used in smear campaigns.
In the United States, AI was deployed in an election-related robocall that mimicked President Joe Biden’s voice, misleading voters in New Hampshire.
As ByteDance advances its AI capabilities, the U.S. is investing heavily in artificial intelligence to compete with China’s technological growth. Last month, former President Donald Trump announced a $500 billion private-sector AI initiative, bringing together OpenAI, SoftBank, and Oracle to develop cutting-edge AI technologies.
However, Cohen believes the U.S. government remains unprepared for the potential security risks posed by AI.
“Our federal agencies have been slow to address these threats,” Cohen said. “Until we take more proactive steps, we’ll be playing catch-up against emerging AI-driven dangers.”
While ByteDance has yet to release OmniHuman-1 for public use, experts predict it could soon be integrated into TikTok and other platforms, adding to the ongoing debate over the app’s security risks.
