ByteDance, the parent company of TikTok, has introduced a groundbreaking AI system called OmniHuman-1, which may be the most realistic deepfake generator to date.
Deepfake technology is widely available, with numerous apps capable of inserting individuals into images or making them appear to say things they never did. However, many deepfake videos still struggle with realism, often exhibiting telltale signs of AI manipulation.
OmniHuman-1 appears to overcome this hurdle, at least based on the carefully selected examples provided by ByteDance’s research team.
One particularly striking demonstration is a fictional performance by Taylor Swift. According to ByteDance researchers, OmniHuman-1 requires only a single reference image and an audio clip—such as speech or singing—to create a video of any desired length. Users can adjust the video's aspect ratio and modify the subject’s body proportions to control how much of their form appears in the footage. The system, trained on 19,000 hours of video content from undisclosed sources, can even alter existing videos, including modifying a person’s limb movements. The results can be remarkably convincing.
Nevertheless, OmniHuman-1 is not without flaws. ByteDance acknowledges that low-quality reference images degrade the final output, and the AI struggles with certain complex poses. Some movements, such as unnatural gestures in a demonstration video, reveal minor imperfections. Despite these limitations, OmniHuman-1 represents a significant leap forward in deepfake realism and may set the stage for future advancements. Although ByteDance has not released the system, similar models could be reverse-engineered by the AI community before long.
The potential consequences of this technology are concerning.
In recent years, deepfake videos have been widely used for political deception. On election day in Taiwan, a group linked to the Chinese Communist Party disseminated AI-generated, misleading audio of a politician endorsing a pro-China candidate. In Moldova, deepfake footage falsely depicted President Maia Sandu resigning. Similarly, South Africa saw deepfake content of rapper Eminem appearing to support an opposition party before the country’s election.
Deepfakes are also becoming a tool for financial crimes. Scammers exploit AI-generated content to impersonate celebrities in fraudulent investment schemes, while businesses are being tricked into transferring vast sums of money to deepfake fraudsters. According to Deloitte, AI-driven fraud accounted for more than $12 billion in losses in 2023, with projections estimating this could reach $40 billion in the U.S. by 2027.
In response, hundreds of AI experts signed an open letter last year calling for stricter deepfake regulations. While there is no federal law criminalizing deepfakes in the U.S., over ten states have enacted legislation against AI-assisted impersonation. California’s proposed law, currently stalled, would be the first to grant judges the authority to order the removal of deepfake content and impose financial penalties on violators.
Despite these efforts, detecting deepfakes remains challenging. Although some social media platforms and search engines have implemented measures to curb their spread, deepfake content continues to proliferate at an alarming rate.
A May 2024 survey conducted by ID verification firm Jumio found that 60% of respondents had encountered a deepfake in the past year. Seventy-two percent expressed concern about being deceived by deepfakes daily, and a majority supported legislative action to combat the rise of AI-generated forgeries.
Validate your login
Sign In
Create New Account