Deepfake videos are becoming alarmingly realistic, and a new AI system from ByteDance, called OmniHuman-1, is pushing the boundaries even further. This cutting-edge tech can produce hyper-convincing fake videos that could easily fool the untrained eye.

Unlike older deepfake tools, OmniHuman-1 only needs a single image and audio clip to create a video. Want to see Taylor Swift perform a song she never sang? Done. How about a TED Talk Einstein never gave? Easy. The system can even tweak body proportions and edit existing videos, making it incredibly versatile.

However, it’s not flawless. Low-quality reference images can lead to less convincing results, and certain poses still trip up the AI. For instance, one demo showed odd hand movements while holding a wine glass. Still, experts agree this is a massive leap forward in deepfake technology.

The implications are troubling. Deepfakes have already been used to spread political misinformation, commit financial fraud, and manipulate public opinion. Last year alone, AI-generated content contributed to over $12 billion in fraud losses, and that number could skyrocket to $40 billion by 2027.

While some states in the U.S. have started to regulate deepfakes, efforts to curb their misuse are still playing catch-up. Social media platforms are trying to limit their spread, but detecting fakes remains a challenge.

A recent survey found that 60% of people encountered a deepfake in the past year, and 72% worry about being fooled daily. Public support for stricter laws is growing, but until then, the rapid rise of deepfakes like those from OmniHuman-1 is a clear call for vigilance.

Stay sharp—these AI creations might look real, but they’re not always what they seem.